Share and discover more about AI with social posts from the community.huggingface/OpenAi
Automatic Prompt Engineer (APE)
APE
Image Source: Zhou et al., (2022)
Zhou et al., (2022) propose automatic prompt engineer (APE) a framework for automatic instruction generation and selection. The instruction generation problem is framed as natural language synthesis addressed as a black-box optimization problem using LLMs to generate and search over candidate solutions.
The first step involves a large language model (as an inference model) that is given output demonstrations to generate instruction candidates for a task. These candidate solutions will guide the search procedure. The instructions are executed using a target model, and then the most appropriate instruction is selected based on computed evaluation scores.
APE discovers a better zero-shot CoT prompt than the human engineered "Let's think step by step" prompt (Kojima et al., 2022).
The prompt "Let's work this out in a step by step way to be sure we have the right answer." elicits chain-of-thought reasoning and improves performance on the MultiArith and GSM8K benchmarks:
APECOT
Image Source: Zhou et al., (2022)
This paper touches on an important topic related to prompt engineering which is the idea of automatically optimizing prompts. While we don't go deep into this topic in this guide, here are a few key papers if you are interested in the topic:
Prompt-OIRL - proposes to use offline inverse reinforcement learning to generate query-dependent prompts.
OPRO - introduces the idea of using LLMs to optimize prompts: let LLMs "Take a deep breath" improves the performance on math problems.
AutoPrompt - proposes an approach to automatically create prompts for a diverse set of tasks based on gradient-guided search.
Prefix Tuning - a lightweight alternative to fine-tuning that prepends a trainable continuous prefix for NLG tasks.
Prompt Tuning - proposes a mechanism for learning soft prompts through backpropagation.https://www.promptingguide.ai/techniques/ape
APE
Image Source: Zhou et al., (2022)
Zhou et al., (2022) propose automatic prompt engineer (APE) a framework for automatic instruction generation and selection. The instruction generation problem is framed as natural language synthesis addressed as a black-box optimization problem using LLMs to generate and search over candidate solutions.
The first step involves a large language model (as an inference model) that is given output demonstrations to generate instruction candidates for a task. These candidate solutions will guide the search procedure. The instructions are executed using a target model, and then the most appropriate instruction is selected based on computed evaluation scores.
APE discovers a better zero-shot CoT prompt than the human engineered "Let's think step by step" prompt (Kojima et al., 2022).
The prompt "Let's work this out in a step by step way to be sure we have the right answer." elicits chain-of-thought reasoning and improves performance on the MultiArith and GSM8K benchmarks:
APECOT
Image Source: Zhou et al., (2022)
This paper touches on an important topic related to prompt engineering which is the idea of automatically optimizing prompts. While we don't go deep into this topic in this guide, here are a few key papers if you are interested in the topic:
Prompt-OIRL - proposes to use offline inverse reinforcement learning to generate query-dependent prompts.
OPRO - introduces the idea of using LLMs to optimize prompts: let LLMs "Take a deep breath" improves the performance on math problems.
AutoPrompt - proposes an approach to automatically create prompts for a diverse set of tasks based on gradient-guided search.
Prefix Tuning - a lightweight alternative to fine-tuning that prepends a trainable continuous prefix for NLG tasks.
Prompt Tuning - proposes a mechanism for learning soft prompts through backpropagation.https://www.promptingguide.ai/techniques/ape
Automatic Reasoning and Tool-use (ART)
Combining CoT prompting and tools in an interleaved manner has shown to be a strong and robust approach to address many tasks with LLMs. These approaches typically require hand-crafting task-specific demonstrations and carefully scripted interleaving of model generations with tool use. Paranjape et al., (2023) propose a new framework that uses a frozen LLM to automatically generate intermediate reasoning steps as a program.
ART works as follows:
given a new task, it select demonstrations of multi-step reasoning and tool use from a task library
at test time, it pauses generation whenever external tools are called, and integrate their output before resuming generation
ART encourages the model to generalize from demonstrations to decompose a new task and use tools in appropriate places, in a zero-shot fashion. In addition, ART is extensible as it also enables humans to fix mistakes in the reasoning steps or add new tools by simply updating the task and tool libraries. The process is demonstrated below:
ART
Image Source: Paranjape et al., (2023)
ART substantially improves over few-shot prompting and automatic CoT on unseen tasks in the BigBench and MMLU benchmarks, and exceeds performance of hand-crafted CoT prompts when human feedback is incorporated.
Below is a table demonstrating ART's performance on BigBench and MMLU tasks:
ART2
Image Source: Paranjape et al., (2023) https://www.promptingguide.ai/techniques/art
Combining CoT prompting and tools in an interleaved manner has shown to be a strong and robust approach to address many tasks with LLMs. These approaches typically require hand-crafting task-specific demonstrations and carefully scripted interleaving of model generations with tool use. Paranjape et al., (2023) propose a new framework that uses a frozen LLM to automatically generate intermediate reasoning steps as a program.
ART works as follows:
given a new task, it select demonstrations of multi-step reasoning and tool use from a task library
at test time, it pauses generation whenever external tools are called, and integrate their output before resuming generation
ART encourages the model to generalize from demonstrations to decompose a new task and use tools in appropriate places, in a zero-shot fashion. In addition, ART is extensible as it also enables humans to fix mistakes in the reasoning steps or add new tools by simply updating the task and tool libraries. The process is demonstrated below:
ART
Image Source: Paranjape et al., (2023)
ART substantially improves over few-shot prompting and automatic CoT on unseen tasks in the BigBench and MMLU benchmarks, and exceeds performance of hand-crafted CoT prompts when human feedback is incorporated.
Below is a table demonstrating ART's performance on BigBench and MMLU tasks:
ART2
Image Source: Paranjape et al., (2023) https://www.promptingguide.ai/techniques/art
Retrieval Augmented Generation (RAG)
General-purpose language models can be fine-tuned to achieve several common tasks such as sentiment analysis and named entity recognition. These tasks generally don't require additional background knowledge.
For more complex and knowledge-intensive tasks, it's possible to build a language model-based system that accesses external knowledge sources to complete tasks. This enables more factual consistency, improves reliability of the generated responses, and helps to mitigate the problem of "hallucination".
Meta AI researchers introduced a method called Retrieval Augmented Generation (RAG) to address such knowledge-intensive tasks. RAG combines an information retrieval component with a text generator model. RAG can be fine-tuned and its internal knowledge can be modified in an efficient manner and without needing retraining of the entire model.
RAG takes an input and retrieves a set of relevant/supporting documents given a source (e.g., Wikipedia). The documents are concatenated as context with the original input prompt and fed to the text generator which produces the final output. This makes RAG adaptive for situations where facts could evolve over time. This is very useful as LLMs's parametric knowledge is static. RAG allows language models to bypass retraining, enabling access to the latest information for generating reliable outputs via retrieval-based generation.
Lewis et al., (2021) proposed a general-purpose fine-tuning recipe for RAG. A pre-trained seq2seq model is used as the parametric memory and a dense vector index of Wikipedia is used as non-parametric memory (accessed using a neural pre-trained retriever). Below is a overview of how the approach works:
RAG
Image Source: Lewis et el. (2021)
RAG performs strong on several benchmarks such as Natural Questions, WebQuestions, and CuratedTrec. RAG generates responses that are more factual, specific, and diverse when tested on MS-MARCO and Jeopardy questions. RAG also improves results on FEVER fact verification.
This shows the potential of RAG as a viable option for enhancing outputs of language models in knowledge-intensive tasks.
More recently, these retriever-based approaches have become more popular and are combined with popular LLMs like ChatGPT to improve capabilities and factual consistency.
RAG Use Case: Generating Friendly ML Paper Titles
Below, we have prepared a notebook tutorial showcasing the use of open-source LLMs to build a RAG system for generating short and concise machine learning paper titles:https://www.promptingguide.ai/techniques/rag
General-purpose language models can be fine-tuned to achieve several common tasks such as sentiment analysis and named entity recognition. These tasks generally don't require additional background knowledge.
For more complex and knowledge-intensive tasks, it's possible to build a language model-based system that accesses external knowledge sources to complete tasks. This enables more factual consistency, improves reliability of the generated responses, and helps to mitigate the problem of "hallucination".
Meta AI researchers introduced a method called Retrieval Augmented Generation (RAG) to address such knowledge-intensive tasks. RAG combines an information retrieval component with a text generator model. RAG can be fine-tuned and its internal knowledge can be modified in an efficient manner and without needing retraining of the entire model.
RAG takes an input and retrieves a set of relevant/supporting documents given a source (e.g., Wikipedia). The documents are concatenated as context with the original input prompt and fed to the text generator which produces the final output. This makes RAG adaptive for situations where facts could evolve over time. This is very useful as LLMs's parametric knowledge is static. RAG allows language models to bypass retraining, enabling access to the latest information for generating reliable outputs via retrieval-based generation.
Lewis et al., (2021) proposed a general-purpose fine-tuning recipe for RAG. A pre-trained seq2seq model is used as the parametric memory and a dense vector index of Wikipedia is used as non-parametric memory (accessed using a neural pre-trained retriever). Below is a overview of how the approach works:
RAG
Image Source: Lewis et el. (2021)
RAG performs strong on several benchmarks such as Natural Questions, WebQuestions, and CuratedTrec. RAG generates responses that are more factual, specific, and diverse when tested on MS-MARCO and Jeopardy questions. RAG also improves results on FEVER fact verification.
This shows the potential of RAG as a viable option for enhancing outputs of language models in knowledge-intensive tasks.
More recently, these retriever-based approaches have become more popular and are combined with popular LLMs like ChatGPT to improve capabilities and factual consistency.
RAG Use Case: Generating Friendly ML Paper Titles
Below, we have prepared a notebook tutorial showcasing the use of open-source LLMs to build a RAG system for generating short and concise machine learning paper titles:https://www.promptingguide.ai/techniques/rag
Tree of Thoughts (ToT)
For complex tasks that require exploration or strategic lookahead, traditional or simple prompting techniques fall short. Yao et el. (2023) and Long (2023) recently proposed Tree of Thoughts (ToT), a framework that generalizes over chain-of-thought prompting and encourages exploration over thoughts that serve as intermediate steps for general problem solving with language models.
ToT maintains a tree of thoughts, where thoughts represent coherent language sequences that serve as intermediate steps toward solving a problem. This approach enables an LM to self-evaluate the progress through intermediate thoughts made towards solving a problem through a deliberate reasoning process. The LM's ability to generate and evaluate thoughts is then combined with search algorithms (e.g., breadth-first search and depth-first search) to enable systematic exploration of thoughts with lookahead and backtracking.
The ToT framework is illustrated below:
TOT
Image Source: Yao et el. (2023)
When using ToT, different tasks requires defining the number of candidates and the number of thoughts/steps. For instance, as demonstrated in the paper, Game of 24 is used as a mathematical reasoning task which requires decomposing the thoughts into 3 steps, each involving an intermediate equation. At each step, the best b=5 candidates are kept.
To perform BFS in ToT for the Game of 24 task, the LM is prompted to evaluate each thought candidate as "sure/maybe/impossible" with regard to reaching 24. As stated by the authors, "the aim is to promote correct partial solutions that can be verdicted within few lookahead trials, and eliminate impossible partial solutions based on "too big/small" commonsense, and keep the rest "maybe"". Values are sampled 3 times for each thought. The process is illustrated below:
TOT2
Image Source: Yao et el. (2023)
From the results reported in the figure below, ToT substantially outperforms the other prompting methods:
TOT3
Image Source: Yao et el. (2023)
Code available here and here
At a high level, the main ideas of Yao et el. (2023) and Long (2023) are similar. Both enhance LLM's capability for complex problem solving through tree search via a multi-round conversation. One of the main difference is that Yao et el. (2023) leverages DFS/BFS/beam search, while the tree search strategy (i.e. when to backtrack and backtracking by how many levels, etc.) proposed in Long (2023) is driven by a "ToT Controller" trained through reinforcement learning. DFS/BFS/Beam search are generic solution search strategies with no adaptation to specific problems. In comparison, a ToT Controller trained through RL might be able learn from new data set or through self-play (AlphaGo vs brute force search), and hence the RL-based ToT system can continue to evolve and learn new knowledge even with a fixed LLM.
Hulbert (2023) has proposed Tree-of-Thought Prompting, which applies the main concept from ToT frameworks as a simple prompting technique, getting the LLM to evaluate intermediate thoughts in a single prompt. A sample ToT prompt is:
Imagine three different experts are answering this question.
All experts will write down 1 step of their thinking,
then share it with the group.
Then all experts will go on to the next step, etc.
If any expert realises they're wrong at any point then they leave.
The question is...
Sun (2023) benchmarked the Tree-of-Thought Prompting with large-scale experiments, and introduce PanelGPT --- an idea of prompting with Panel discussions among LLMs.https://www.promptingguide.ai/techniques/tot
For complex tasks that require exploration or strategic lookahead, traditional or simple prompting techniques fall short. Yao et el. (2023) and Long (2023) recently proposed Tree of Thoughts (ToT), a framework that generalizes over chain-of-thought prompting and encourages exploration over thoughts that serve as intermediate steps for general problem solving with language models.
ToT maintains a tree of thoughts, where thoughts represent coherent language sequences that serve as intermediate steps toward solving a problem. This approach enables an LM to self-evaluate the progress through intermediate thoughts made towards solving a problem through a deliberate reasoning process. The LM's ability to generate and evaluate thoughts is then combined with search algorithms (e.g., breadth-first search and depth-first search) to enable systematic exploration of thoughts with lookahead and backtracking.
The ToT framework is illustrated below:
TOT
Image Source: Yao et el. (2023)
When using ToT, different tasks requires defining the number of candidates and the number of thoughts/steps. For instance, as demonstrated in the paper, Game of 24 is used as a mathematical reasoning task which requires decomposing the thoughts into 3 steps, each involving an intermediate equation. At each step, the best b=5 candidates are kept.
To perform BFS in ToT for the Game of 24 task, the LM is prompted to evaluate each thought candidate as "sure/maybe/impossible" with regard to reaching 24. As stated by the authors, "the aim is to promote correct partial solutions that can be verdicted within few lookahead trials, and eliminate impossible partial solutions based on "too big/small" commonsense, and keep the rest "maybe"". Values are sampled 3 times for each thought. The process is illustrated below:
TOT2
Image Source: Yao et el. (2023)
From the results reported in the figure below, ToT substantially outperforms the other prompting methods:
TOT3
Image Source: Yao et el. (2023)
Code available here and here
At a high level, the main ideas of Yao et el. (2023) and Long (2023) are similar. Both enhance LLM's capability for complex problem solving through tree search via a multi-round conversation. One of the main difference is that Yao et el. (2023) leverages DFS/BFS/beam search, while the tree search strategy (i.e. when to backtrack and backtracking by how many levels, etc.) proposed in Long (2023) is driven by a "ToT Controller" trained through reinforcement learning. DFS/BFS/Beam search are generic solution search strategies with no adaptation to specific problems. In comparison, a ToT Controller trained through RL might be able learn from new data set or through self-play (AlphaGo vs brute force search), and hence the RL-based ToT system can continue to evolve and learn new knowledge even with a fixed LLM.
Hulbert (2023) has proposed Tree-of-Thought Prompting, which applies the main concept from ToT frameworks as a simple prompting technique, getting the LLM to evaluate intermediate thoughts in a single prompt. A sample ToT prompt is:
Imagine three different experts are answering this question.
All experts will write down 1 step of their thinking,
then share it with the group.
Then all experts will go on to the next step, etc.
If any expert realises they're wrong at any point then they leave.
The question is...
Sun (2023) benchmarked the Tree-of-Thought Prompting with large-scale experiments, and introduce PanelGPT --- an idea of prompting with Panel discussions among LLMs.https://www.promptingguide.ai/techniques/tot
Given a set of relevant quotes (delimited by <quotes></quotes>) extracted from a document and the original document (delimited by ####), please compose an answer to the question. Ensure that the answer is accurate, has a friendly tone, and sounds helpful.
####
{{document}}
####
<quotes>
- Chain-of-thought (CoT) prompting[27]
- Generated knowledge prompting[37]
- Least-to-most prompting[38]
- Self-consistency decoding[39]
- Complexity-based prompting[41]
- Self-refine[42]
- Tree-of-thought prompting[43]
- Maieutic prompting[45]
- Directional-stimulus prompting[46]
- Textual inversion and embeddings[59]
- Using gradient descent to search for prompts[61][62][63][64]
- Prompt injection[65][66][67]
</quotes>
Output of Prompt 2:
The prompting techniques mentioned in the document include:
1. Chain-of-thought (CoT) prompting[27]
2. Generated knowledge prompting[37]
3. Least-to-most prompting[38]
4. Self-consistency decoding[39]
5. Complexity-based prompting[41]
6. Self-refine[42]
7. Tree-of-thought prompting[43]
8. Maieutic prompting[45]
9. Directional-stimulus prompting[46]
10. Textual inversion and embeddings[59]
11. Using gradient descent to search for prompts[61][62][63][64]
12. Prompt injection[65][66][67]
Each of these techniques employs unique strategies to enhance or specify the interactions with large language models to produce the desired outcomes.
As you can see, simplifying and creating prompt chains is a useful prompting approach where the responses need to undergo several operations or transformations. As an exercise, feel free to design a prompt that removes the citations (e.g., [27]) from the response before sending this as a final response to the user of your application.
You can also find more examples of prompt chaining in this documentation that leverages the Claude LLM. Our example is inspired and adapted from their examples.https://www.promptingguide.ai/techniques/prompt_chaining
####
{{document}}
####
<quotes>
- Chain-of-thought (CoT) prompting[27]
- Generated knowledge prompting[37]
- Least-to-most prompting[38]
- Self-consistency decoding[39]
- Complexity-based prompting[41]
- Self-refine[42]
- Tree-of-thought prompting[43]
- Maieutic prompting[45]
- Directional-stimulus prompting[46]
- Textual inversion and embeddings[59]
- Using gradient descent to search for prompts[61][62][63][64]
- Prompt injection[65][66][67]
</quotes>
Output of Prompt 2:
The prompting techniques mentioned in the document include:
1. Chain-of-thought (CoT) prompting[27]
2. Generated knowledge prompting[37]
3. Least-to-most prompting[38]
4. Self-consistency decoding[39]
5. Complexity-based prompting[41]
6. Self-refine[42]
7. Tree-of-thought prompting[43]
8. Maieutic prompting[45]
9. Directional-stimulus prompting[46]
10. Textual inversion and embeddings[59]
11. Using gradient descent to search for prompts[61][62][63][64]
12. Prompt injection[65][66][67]
Each of these techniques employs unique strategies to enhance or specify the interactions with large language models to produce the desired outcomes.
As you can see, simplifying and creating prompt chains is a useful prompting approach where the responses need to undergo several operations or transformations. As an exercise, feel free to design a prompt that removes the citations (e.g., [27]) from the response before sending this as a final response to the user of your application.
You can also find more examples of prompt chaining in this documentation that leverages the Claude LLM. Our example is inspired and adapted from their examples.https://www.promptingguide.ai/techniques/prompt_chaining
Prompt Chaining
Introduction to Prompt Chaining
To improve the reliability and performance of LLMs, one of the important prompt engineering techniques is to break tasks into its subtasks. Once those subtasks have been identified, the LLM is prompted with a subtask and then its response is used as input to another prompt. This is what's referred to as prompt chaining, where a task is split into subtasks with the idea to create a chain of prompt operations.
Prompt chaining is useful to accomplish complex tasks which an LLM might struggle to address if prompted with a very detailed prompt. In prompt chaining, chain prompts perform transformations or additional processes on the generated responses before reaching a final desired state.
Besides achieving better performance, prompt chaining helps to boost the transparency of your LLM application, increases controllability, and reliability. This means that you can debug problems with model responses much more easily and analyze and improve performance in the different stages that need improvement.
Prompt chaining is particularly useful when building LLM-powered conversational assistants and improving the personalization and user experience of your applications.
Use Cases for Prompt Chaining
Prompt Chaining for Document QA
Prompt chaining can be used in different scenarios that could involve several operations or transformations. For instance, one common use case of LLMs involves answering questions about a large text document. It helps if you design two different prompts where the first prompt is responsible for extracting relevant quotes to answer a question and a second prompt takes as input the quotes and original document to answer a given question. In other words, you will be creating two different prompts to perform the task of answering a question given in a document.
The first prompt below extracts the relevant quotes from the document given the question. Note that for simplicity, we have added a placeholder for the document {{document}}. To test the prompt you can copy and paste an article from Wikipedia such as this page for prompt engineering. Due to larger context used for this task, we are using the gpt-4-1106-preview model from OpenAI. You can use the prompt with other long-context LLMs like Claude.
Prompt 1:
You are a helpful assistant. Your task is to help answer a question given in a document. The first step is to extract quotes relevant to the question from the document, delimited by ####. Please output the list of quotes using <quotes></quotes>. Respond with "No relevant quotes found!" if no relevant quotes were found.
####
{{document}}
####
Here is a screenshot of the entire prompt including the question which is passed using the user role.
Prompt Chaining Part 1
Output of Prompt 1:
<quotes>
- Chain-of-thought (CoT) prompting[27]
- Generated knowledge prompting[37]
- Least-to-most prompting[38]
- Self-consistency decoding[39]
- Complexity-based prompting[41]
- Self-refine[42]
- Tree-of-thought prompting[43]
- Maieutic prompting[45]
- Directional-stimulus prompting[46]
- Textual inversion and embeddings[59]
- Using gradient descent to search for prompts[61][62][63][64]
- Prompt injection[65][66][67]
</quotes>
The quotes that were returned in the first prompt can now be used as input to the second prompt below. Note that you can clean up the quotes a bit more, i.e., remove the citations. Those citations could be removed or utilized as part of another prompt in the chain but you can ignore this for now. The second prompt then takes the relevant quotes extracted by prompt 1 and prepares a helpful response to the question given in the document and those extracted quotes. The second prompt can be the following:
Prompt 2:
Introduction to Prompt Chaining
To improve the reliability and performance of LLMs, one of the important prompt engineering techniques is to break tasks into its subtasks. Once those subtasks have been identified, the LLM is prompted with a subtask and then its response is used as input to another prompt. This is what's referred to as prompt chaining, where a task is split into subtasks with the idea to create a chain of prompt operations.
Prompt chaining is useful to accomplish complex tasks which an LLM might struggle to address if prompted with a very detailed prompt. In prompt chaining, chain prompts perform transformations or additional processes on the generated responses before reaching a final desired state.
Besides achieving better performance, prompt chaining helps to boost the transparency of your LLM application, increases controllability, and reliability. This means that you can debug problems with model responses much more easily and analyze and improve performance in the different stages that need improvement.
Prompt chaining is particularly useful when building LLM-powered conversational assistants and improving the personalization and user experience of your applications.
Use Cases for Prompt Chaining
Prompt Chaining for Document QA
Prompt chaining can be used in different scenarios that could involve several operations or transformations. For instance, one common use case of LLMs involves answering questions about a large text document. It helps if you design two different prompts where the first prompt is responsible for extracting relevant quotes to answer a question and a second prompt takes as input the quotes and original document to answer a given question. In other words, you will be creating two different prompts to perform the task of answering a question given in a document.
The first prompt below extracts the relevant quotes from the document given the question. Note that for simplicity, we have added a placeholder for the document {{document}}. To test the prompt you can copy and paste an article from Wikipedia such as this page for prompt engineering. Due to larger context used for this task, we are using the gpt-4-1106-preview model from OpenAI. You can use the prompt with other long-context LLMs like Claude.
Prompt 1:
You are a helpful assistant. Your task is to help answer a question given in a document. The first step is to extract quotes relevant to the question from the document, delimited by ####. Please output the list of quotes using <quotes></quotes>. Respond with "No relevant quotes found!" if no relevant quotes were found.
####
{{document}}
####
Here is a screenshot of the entire prompt including the question which is passed using the user role.
Prompt Chaining Part 1
Output of Prompt 1:
<quotes>
- Chain-of-thought (CoT) prompting[27]
- Generated knowledge prompting[37]
- Least-to-most prompting[38]
- Self-consistency decoding[39]
- Complexity-based prompting[41]
- Self-refine[42]
- Tree-of-thought prompting[43]
- Maieutic prompting[45]
- Directional-stimulus prompting[46]
- Textual inversion and embeddings[59]
- Using gradient descent to search for prompts[61][62][63][64]
- Prompt injection[65][66][67]
</quotes>
The quotes that were returned in the first prompt can now be used as input to the second prompt below. Note that you can clean up the quotes a bit more, i.e., remove the citations. Those citations could be removed or utilized as part of another prompt in the chain but you can ignore this for now. The second prompt then takes the relevant quotes extracted by prompt 1 and prepares a helpful response to the question given in the document and those extracted quotes. The second prompt can be the following:
Prompt 2:
No, the objective of golf is not to get a higher point total than others. Rather, the objective is to play a set of holes in the least number of strokes. The total number of strokes is used to determine the winner of the game, not the total number of points.
Question: Part of golf is trying to get a higher point total than others. Yes or No?
Knowledge: Golf is a precision club-and-ball sport in which competing players (or golfers) use many types of clubs to hit balls into a series of holes on a course using the fewest number of strokes. The goal is to complete the course with the lowest score, which is calculated by adding up the total number of strokes taken on each hole. The player with the lowest score wins the game.
Explain and Answer:
Answer 2 (confidence is a lot lower):
Yes, part of golf is trying to get a higher point total than others. Each player tries to complete the course with the lowest score, which is calculated by adding up the total number of strokes taken on each hole. The player with the lowest score wins the game.
Some really interesting things happened with this example. In the first answer, the model was very confident but in the second not so much. I simplified the process for demonstration purposes but there are a few more details to consider when arriving at the final answer. Check out the paper for more.https://www.promptingguide.ai/techniques/knowledge
Question: Part of golf is trying to get a higher point total than others. Yes or No?
Knowledge: Golf is a precision club-and-ball sport in which competing players (or golfers) use many types of clubs to hit balls into a series of holes on a course using the fewest number of strokes. The goal is to complete the course with the lowest score, which is calculated by adding up the total number of strokes taken on each hole. The player with the lowest score wins the game.
Explain and Answer:
Answer 2 (confidence is a lot lower):
Yes, part of golf is trying to get a higher point total than others. Each player tries to complete the course with the lowest score, which is calculated by adding up the total number of strokes taken on each hole. The player with the lowest score wins the game.
Some really interesting things happened with this example. In the first answer, the model was very confident but in the second not so much. I simplified the process for demonstration purposes but there are a few more details to consider when arriving at the final answer. Check out the paper for more.https://www.promptingguide.ai/techniques/knowledge
Generated Knowledge Prompting
GENKNOW
Image Source: Liu et al. 2022
LLMs continue to be improved and one popular technique includes the ability to incorporate knowledge or information to help the model make more accurate predictions.
Using a similar idea, can the model also be used to generate knowledge before making a prediction? That's what is attempted in the paper by Liu et al. 2022 -- generate knowledge to be used as part of the prompt. In particular, how helpful is this for tasks such as commonsense reasoning?
Let's try a simple prompt:
Prompt:
Part of golf is trying to get a higher point total than others. Yes or No?
Output:
Yes.
This type of mistake reveals the limitations of LLMs to perform tasks that require more knowledge about the world. How do we improve this with knowledge generation?
First, we generate a few "knowledges":
Prompt:
Input: Greece is larger than mexico.
Knowledge: Greece is approximately 131,957 sq km, while Mexico is approximately 1,964,375 sq km, making Mexico 1,389% larger than Greece.
Input: Glasses always fog up.
Knowledge: Condensation occurs on eyeglass lenses when water vapor from your sweat, breath, and ambient humidity lands on a cold surface, cools, and then changes into tiny drops of liquid, forming a film that you see as fog. Your lenses will be relatively cool compared to your breath, especially when the outside air is cold.
Input: A fish is capable of thinking.
Knowledge: Fish are more intelligent than they appear. In many areas, such as memory, their cognitive powers match or exceed those of ’higher’ vertebrates including non-human primates. Fish’s long-term memories help them keep track of complex social relationships.
Input: A common effect of smoking lots of cigarettes in one’s lifetime is a higher than normal chance of getting lung cancer.
Knowledge: Those who consistently averaged less than one cigarette per day over their lifetime had nine times the risk of dying from lung cancer than never smokers. Among people who smoked between one and 10 cigarettes per day, the risk of dying from lung cancer was nearly 12 times higher than that of never smokers.
Input: A rock is the same size as a pebble.
Knowledge: A pebble is a clast of rock with a particle size of 4 to 64 millimetres based on the Udden-Wentworth scale of sedimentology. Pebbles are generally considered larger than granules (2 to 4 millimetres diameter) and smaller than cobbles (64 to 256 millimetres diameter).
Input: Part of golf is trying to get a higher point total than others.
Knowledge:
Knowledge 1:
The objective of golf is to play a set of holes in the least number of strokes. A round of golf typically consists of 18 holes. Each hole is played once in the round on a standard golf course. Each stroke is counted as one point, and the total number of strokes is used to determine the winner of the game.
Knowledge 2:
Golf is a precision club-and-ball sport in which competing players (or golfers) use many types of clubs to hit balls into a series of holes on a course using the fewest number of strokes. The goal is to complete the course with the lowest score, which is calculated by adding up the total number of strokes taken on each hole. The player with the lowest score wins the game.
We are using the prompt provided in the paper by Liu et al. 2022.
The next step is to integrate the knowledge and get a prediction. I reformatted the question into QA format to guide the answer format.
Prompt:
Question: Part of golf is trying to get a higher point total than others. Yes or No?
Knowledge: The objective of golf is to play a set of holes in the least number of strokes. A round of golf typically consists of 18 holes. Each hole is played once in the round on a standard golf course. Each stroke is counted as one point, and the total number of strokes is used to determine the winner of the game.
Explain and Answer:
Answer 1 (confidence very high):
GENKNOW
Image Source: Liu et al. 2022
LLMs continue to be improved and one popular technique includes the ability to incorporate knowledge or information to help the model make more accurate predictions.
Using a similar idea, can the model also be used to generate knowledge before making a prediction? That's what is attempted in the paper by Liu et al. 2022 -- generate knowledge to be used as part of the prompt. In particular, how helpful is this for tasks such as commonsense reasoning?
Let's try a simple prompt:
Prompt:
Part of golf is trying to get a higher point total than others. Yes or No?
Output:
Yes.
This type of mistake reveals the limitations of LLMs to perform tasks that require more knowledge about the world. How do we improve this with knowledge generation?
First, we generate a few "knowledges":
Prompt:
Input: Greece is larger than mexico.
Knowledge: Greece is approximately 131,957 sq km, while Mexico is approximately 1,964,375 sq km, making Mexico 1,389% larger than Greece.
Input: Glasses always fog up.
Knowledge: Condensation occurs on eyeglass lenses when water vapor from your sweat, breath, and ambient humidity lands on a cold surface, cools, and then changes into tiny drops of liquid, forming a film that you see as fog. Your lenses will be relatively cool compared to your breath, especially when the outside air is cold.
Input: A fish is capable of thinking.
Knowledge: Fish are more intelligent than they appear. In many areas, such as memory, their cognitive powers match or exceed those of ’higher’ vertebrates including non-human primates. Fish’s long-term memories help them keep track of complex social relationships.
Input: A common effect of smoking lots of cigarettes in one’s lifetime is a higher than normal chance of getting lung cancer.
Knowledge: Those who consistently averaged less than one cigarette per day over their lifetime had nine times the risk of dying from lung cancer than never smokers. Among people who smoked between one and 10 cigarettes per day, the risk of dying from lung cancer was nearly 12 times higher than that of never smokers.
Input: A rock is the same size as a pebble.
Knowledge: A pebble is a clast of rock with a particle size of 4 to 64 millimetres based on the Udden-Wentworth scale of sedimentology. Pebbles are generally considered larger than granules (2 to 4 millimetres diameter) and smaller than cobbles (64 to 256 millimetres diameter).
Input: Part of golf is trying to get a higher point total than others.
Knowledge:
Knowledge 1:
The objective of golf is to play a set of holes in the least number of strokes. A round of golf typically consists of 18 holes. Each hole is played once in the round on a standard golf course. Each stroke is counted as one point, and the total number of strokes is used to determine the winner of the game.
Knowledge 2:
Golf is a precision club-and-ball sport in which competing players (or golfers) use many types of clubs to hit balls into a series of holes on a course using the fewest number of strokes. The goal is to complete the course with the lowest score, which is calculated by adding up the total number of strokes taken on each hole. The player with the lowest score wins the game.
We are using the prompt provided in the paper by Liu et al. 2022.
The next step is to integrate the knowledge and get a prediction. I reformatted the question into QA format to guide the answer format.
Prompt:
Question: Part of golf is trying to get a higher point total than others. Yes or No?
Knowledge: The objective of golf is to play a set of holes in the least number of strokes. A round of golf typically consists of 18 holes. Each hole is played once in the round on a standard golf course. Each stroke is counted as one point, and the total number of strokes is used to determine the winner of the game.
Explain and Answer:
Answer 1 (confidence very high):
Self-Consistency
Perhaps one of the more advanced techniques out there for prompt engineering is self-consistency. Proposed by Wang et al. (2022), self-consistency aims "to replace the naive greedy decoding used in chain-of-thought prompting". The idea is to sample multiple, diverse reasoning paths through few-shot CoT, and use the generations to select the most consistent answer. This helps to boost the performance of CoT prompting on tasks involving arithmetic and commonsense reasoning.
Let's try the following example for arithmetic reasoning:
Prompt:
When I was 6 my sister was half my age. Now
I’m 70 how old is my sister?
Output:
35
The output is wrong! How may we improve this with self-consistency? Let's try it out. We will use the few-shot exemplars from Wang et al. 2022 (Table 17):
Prompt:
Q: There are 15 trees in the grove. Grove workers will plant trees in the grove today. After they are done,
there will be 21 trees. How many trees did the grove workers plant today?
A: We start with 15 trees. Later we have 21 trees. The difference must be the number of trees they planted.
So, they must have planted 21 - 15 = 6 trees. The answer is 6.
Q: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?
A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5.
Q: Leah had 32 chocolates and her sister had 42. If they ate 35, how many pieces do they have left in total?
A: Leah had 32 chocolates and Leah’s sister had 42. That means there were originally 32 + 42 = 74
chocolates. 35 have been eaten. So in total they still have 74 - 35 = 39 chocolates. The answer is 39.
Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops
did Jason give to Denny?
A: Jason had 20 lollipops. Since he only has 12 now, he must have given the rest to Denny. The number of
lollipops he has given to Denny must have been 20 - 12 = 8 lollipops. The answer is 8.
Q: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does
he have now?
A: He has 5 toys. He got 2 from mom, so after that he has 5 + 2 = 7 toys. Then he got 2 more from dad, so
in total he has 7 + 2 = 9 toys. The answer is 9.
Q: There were nine computers in the server room. Five more computers were installed each day, from
monday to thursday. How many computers are now in the server room?
A: There are 4 days from monday to thursday. 5 computers were added each day. That means in total 4 * 5 =
20 computers were added. There were 9 computers in the beginning, so now there are 9 + 20 = 29 computers.
The answer is 29.
Q: Michael had 58 golf balls. On tuesday, he lost 23 golf balls. On wednesday, he lost 2 more. How many
golf balls did he have at the end of wednesday?
A: Michael initially had 58 balls. He lost 23 on Tuesday, so after that he has 58 - 23 = 35 balls. On
Wednesday he lost 2 more so now he has 35 - 2 = 33 balls. The answer is 33.
Q: Olivia has $23. She bought five bagels for $3 each. How much money does she have left?
A: She bought 5 bagels for $3 each. This means she spent $15. She has $8 left.
Q: When I was 6 my sister was half my age. Now I’m 70 how old is my sister?
A:
Output 1:
When I was 6 my sister was half my age, so she was 3. Now I am 70, so she is 70 - 3 = 67. The answer is 67.
Output 2:
When the narrator was 6, his sister was half his age, which is 3. Now that the narrator is 70, his sister would be 70 - 3 = 67 years old. The answer is 67.
Output 3:
When I was 6 my sister was half my age, so she was 3. Now I am 70, so she is 70/2 = 35. The answer is 35.
Computing for the final answer involves a few steps (check out the paper for the details) but for the sake of simplicity, we can see that there is already a majority answer emerging so that would essentially become the final answer.https://www.promptingguide.ai/techniques/consistency
Perhaps one of the more advanced techniques out there for prompt engineering is self-consistency. Proposed by Wang et al. (2022), self-consistency aims "to replace the naive greedy decoding used in chain-of-thought prompting". The idea is to sample multiple, diverse reasoning paths through few-shot CoT, and use the generations to select the most consistent answer. This helps to boost the performance of CoT prompting on tasks involving arithmetic and commonsense reasoning.
Let's try the following example for arithmetic reasoning:
Prompt:
When I was 6 my sister was half my age. Now
I’m 70 how old is my sister?
Output:
35
The output is wrong! How may we improve this with self-consistency? Let's try it out. We will use the few-shot exemplars from Wang et al. 2022 (Table 17):
Prompt:
Q: There are 15 trees in the grove. Grove workers will plant trees in the grove today. After they are done,
there will be 21 trees. How many trees did the grove workers plant today?
A: We start with 15 trees. Later we have 21 trees. The difference must be the number of trees they planted.
So, they must have planted 21 - 15 = 6 trees. The answer is 6.
Q: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?
A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5.
Q: Leah had 32 chocolates and her sister had 42. If they ate 35, how many pieces do they have left in total?
A: Leah had 32 chocolates and Leah’s sister had 42. That means there were originally 32 + 42 = 74
chocolates. 35 have been eaten. So in total they still have 74 - 35 = 39 chocolates. The answer is 39.
Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops
did Jason give to Denny?
A: Jason had 20 lollipops. Since he only has 12 now, he must have given the rest to Denny. The number of
lollipops he has given to Denny must have been 20 - 12 = 8 lollipops. The answer is 8.
Q: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does
he have now?
A: He has 5 toys. He got 2 from mom, so after that he has 5 + 2 = 7 toys. Then he got 2 more from dad, so
in total he has 7 + 2 = 9 toys. The answer is 9.
Q: There were nine computers in the server room. Five more computers were installed each day, from
monday to thursday. How many computers are now in the server room?
A: There are 4 days from monday to thursday. 5 computers were added each day. That means in total 4 * 5 =
20 computers were added. There were 9 computers in the beginning, so now there are 9 + 20 = 29 computers.
The answer is 29.
Q: Michael had 58 golf balls. On tuesday, he lost 23 golf balls. On wednesday, he lost 2 more. How many
golf balls did he have at the end of wednesday?
A: Michael initially had 58 balls. He lost 23 on Tuesday, so after that he has 58 - 23 = 35 balls. On
Wednesday he lost 2 more so now he has 35 - 2 = 33 balls. The answer is 33.
Q: Olivia has $23. She bought five bagels for $3 each. How much money does she have left?
A: She bought 5 bagels for $3 each. This means she spent $15. She has $8 left.
Q: When I was 6 my sister was half my age. Now I’m 70 how old is my sister?
A:
Output 1:
When I was 6 my sister was half my age, so she was 3. Now I am 70, so she is 70 - 3 = 67. The answer is 67.
Output 2:
When the narrator was 6, his sister was half his age, which is 3. Now that the narrator is 70, his sister would be 70 - 3 = 67 years old. The answer is 67.
Output 3:
When I was 6 my sister was half my age, so she was 3. Now I am 70, so she is 70/2 = 35. The answer is 35.
Computing for the final answer involves a few steps (check out the paper for the details) but for the sake of simplicity, we can see that there is already a majority answer emerging so that would essentially become the final answer.https://www.promptingguide.ai/techniques/consistency
Stage 1): question clustering: partition questions of a given dataset into a few clusters
Stage 2): demonstration sampling: select a representative question from each cluster and generate its reasoning chain using Zero-Shot-CoT with simple heuristics
The simple heuristics could be length of questions (e.g., 60 tokens) and number of steps in rationale (e.g., 5 reasoning steps). This encourages the model to use simple and accurate demonstrations.
The process is illustrated below:
AUTOCOT
Image Source: Zhang et al. (2022)
Code for Auto-CoT is available here.https://www.promptingguide.ai/techniques/cot
Stage 2): demonstration sampling: select a representative question from each cluster and generate its reasoning chain using Zero-Shot-CoT with simple heuristics
The simple heuristics could be length of questions (e.g., 60 tokens) and number of steps in rationale (e.g., 5 reasoning steps). This encourages the model to use simple and accurate demonstrations.
The process is illustrated below:
AUTOCOT
Image Source: Zhang et al. (2022)
Code for Auto-CoT is available here.https://www.promptingguide.ai/techniques/cot
Chain-of-Thought Prompting
Chain-of-Thought (CoT) Prompting
COT
Image Source: Wei et al. (2022)
Introduced in Wei et al. (2022), chain-of-thought (CoT) prompting enables complex reasoning capabilities through intermediate reasoning steps. You can combine it with few-shot prompting to get better results on more complex tasks that require reasoning before responding.
Prompt:
The odd numbers in this group add up to an even number: 4, 8, 9, 15, 12, 2, 1.
A: Adding all the odd numbers (9, 15, 1) gives 25. The answer is False.
The odd numbers in this group add up to an even number: 17, 10, 19, 4, 8, 12, 24.
A: Adding all the odd numbers (17, 19) gives 36. The answer is True.
The odd numbers in this group add up to an even number: 16, 11, 14, 4, 8, 13, 24.
A: Adding all the odd numbers (11, 13) gives 24. The answer is True.
The odd numbers in this group add up to an even number: 17, 9, 10, 12, 13, 4, 2.
A: Adding all the odd numbers (17, 9, 13) gives 39. The answer is False.
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
Output:
Adding all the odd numbers (15, 5, 13, 7, 1) gives 41. The answer is False.
Wow! We can see a perfect result when we provided the reasoning step. In fact, we can solve this task by providing even fewer examples, i.e., just one example seems enough:
Prompt:
The odd numbers in this group add up to an even number: 4, 8, 9, 15, 12, 2, 1.
A: Adding all the odd numbers (9, 15, 1) gives 25. The answer is False.
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
Output:
Adding all the odd numbers (15, 5, 13, 7, 1) gives 41. The answer is False.
Keep in mind that the authors claim that this is an emergent ability that arises with sufficiently large language models.
Zero-shot COT Prompting
Zero-shot COT
Image Source: Kojima et al. (2022)
One recent idea that came out more recently is the idea of zero-shot CoT (Kojima et al. 2022) that essentially involves adding "Let's think step by step" to the original prompt. Let's try a simple problem and see how the model performs:
Prompt:
I went to the market and bought 10 apples. I gave 2 apples to the neighbor and 2 to the repairman. I then went and bought 5 more apples and ate 1. How many apples did I remain with?
Output:
11 apples
The answer is incorrect! Now Let's try with the special prompt.
Prompt:
I went to the market and bought 10 apples. I gave 2 apples to the neighbor and 2 to the repairman. I then went and bought 5 more apples and ate 1. How many apples did I remain with?
Let's think step by step.
Output:
First, you started with 10 apples.
You gave away 2 apples to the neighbor and 2 to the repairman, so you had 6 apples left.
Then you bought 5 more apples, so now you had 11 apples.
Finally, you ate 1 apple, so you would remain with 10 apples.
It's impressive that this simple prompt is effective at this task. This is particularly useful where you don't have too many examples to use in the prompt.
Want to learn more about advanced use cases of Chain-of-Thought? Check out our new cohort-based course. Use promo code MAVENAI20 for a 20% discount.
Automatic Chain-of-Thought (Auto-CoT)
When applying chain-of-thought prompting with demonstrations, the process involves hand-crafting effective and diverse examples. This manual effort could lead to suboptimal solutions. Zhang et al. (2022) propose an approach to eliminate manual efforts by leveraging LLMs with "Let's think step by step" prompt to generate reasoning chains for demonstrations one by one. This automatic process can still end up with mistakes in generated chains. To mitigate the effects of the mistakes, the diversity of demonstrations matter. This work proposes Auto-CoT, which samples questions with diversity and generates reasoning chains to construct the demonstrations.
Auto-CoT consists of two main stages:
Chain-of-Thought (CoT) Prompting
COT
Image Source: Wei et al. (2022)
Introduced in Wei et al. (2022), chain-of-thought (CoT) prompting enables complex reasoning capabilities through intermediate reasoning steps. You can combine it with few-shot prompting to get better results on more complex tasks that require reasoning before responding.
Prompt:
The odd numbers in this group add up to an even number: 4, 8, 9, 15, 12, 2, 1.
A: Adding all the odd numbers (9, 15, 1) gives 25. The answer is False.
The odd numbers in this group add up to an even number: 17, 10, 19, 4, 8, 12, 24.
A: Adding all the odd numbers (17, 19) gives 36. The answer is True.
The odd numbers in this group add up to an even number: 16, 11, 14, 4, 8, 13, 24.
A: Adding all the odd numbers (11, 13) gives 24. The answer is True.
The odd numbers in this group add up to an even number: 17, 9, 10, 12, 13, 4, 2.
A: Adding all the odd numbers (17, 9, 13) gives 39. The answer is False.
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
Output:
Adding all the odd numbers (15, 5, 13, 7, 1) gives 41. The answer is False.
Wow! We can see a perfect result when we provided the reasoning step. In fact, we can solve this task by providing even fewer examples, i.e., just one example seems enough:
Prompt:
The odd numbers in this group add up to an even number: 4, 8, 9, 15, 12, 2, 1.
A: Adding all the odd numbers (9, 15, 1) gives 25. The answer is False.
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
Output:
Adding all the odd numbers (15, 5, 13, 7, 1) gives 41. The answer is False.
Keep in mind that the authors claim that this is an emergent ability that arises with sufficiently large language models.
Zero-shot COT Prompting
Zero-shot COT
Image Source: Kojima et al. (2022)
One recent idea that came out more recently is the idea of zero-shot CoT (Kojima et al. 2022) that essentially involves adding "Let's think step by step" to the original prompt. Let's try a simple problem and see how the model performs:
Prompt:
I went to the market and bought 10 apples. I gave 2 apples to the neighbor and 2 to the repairman. I then went and bought 5 more apples and ate 1. How many apples did I remain with?
Output:
11 apples
The answer is incorrect! Now Let's try with the special prompt.
Prompt:
I went to the market and bought 10 apples. I gave 2 apples to the neighbor and 2 to the repairman. I then went and bought 5 more apples and ate 1. How many apples did I remain with?
Let's think step by step.
Output:
First, you started with 10 apples.
You gave away 2 apples to the neighbor and 2 to the repairman, so you had 6 apples left.
Then you bought 5 more apples, so now you had 11 apples.
Finally, you ate 1 apple, so you would remain with 10 apples.
It's impressive that this simple prompt is effective at this task. This is particularly useful where you don't have too many examples to use in the prompt.
Want to learn more about advanced use cases of Chain-of-Thought? Check out our new cohort-based course. Use promo code MAVENAI20 for a 20% discount.
Automatic Chain-of-Thought (Auto-CoT)
When applying chain-of-thought prompting with demonstrations, the process involves hand-crafting effective and diverse examples. This manual effort could lead to suboptimal solutions. Zhang et al. (2022) propose an approach to eliminate manual efforts by leveraging LLMs with "Let's think step by step" prompt to generate reasoning chains for demonstrations one by one. This automatic process can still end up with mistakes in generated chains. To mitigate the effects of the mistakes, the diversity of demonstrations matter. This work proposes Auto-CoT, which samples questions with diversity and generates reasoning chains to construct the demonstrations.
Auto-CoT consists of two main stages:
The odd numbers in this group add up to an even number: 4, 8, 9, 15, 12, 2, 1.
A: The answer is False.
The odd numbers in this group add up to an even number: 17, 10, 19, 4, 8, 12, 24.
A: The answer is True.
The odd numbers in this group add up to an even number: 16, 11, 14, 4, 8, 13, 24.
A: The answer is True.
The odd numbers in this group add up to an even number: 17, 9, 10, 12, 13, 4, 2.
A: The answer is False.
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
Output:
The answer is True.
That didn't work. It seems like few-shot prompting is not enough to get reliable responses for this type of reasoning problem. The example above provides basic information on the task. If you take a closer look, the type of task we have introduced involves a few more reasoning steps. In other words, it might help if we break the problem down into steps and demonstrate that to the model. More recently, chain-of-thought (CoT) prompting has been popularized to address more complex arithmetic, commonsense, and symbolic reasoning tasks.
Overall, it seems that providing examples is useful for solving some tasks. When zero-shot prompting and few-shot prompting are not sufficient, it might mean that whatever was learned by the model isn't enough to do well at the task. From here it is recommended to start thinking about fine-tuning your models or experimenting with more advanced prompting techniques. Up next we talk about one of the popular prompting techniques called chain-of-thought prompting which has gained a lot of popularity.
https://www.promptingguide.ai/techniques/fewshot
A: The answer is False.
The odd numbers in this group add up to an even number: 17, 10, 19, 4, 8, 12, 24.
A: The answer is True.
The odd numbers in this group add up to an even number: 16, 11, 14, 4, 8, 13, 24.
A: The answer is True.
The odd numbers in this group add up to an even number: 17, 9, 10, 12, 13, 4, 2.
A: The answer is False.
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
Output:
The answer is True.
That didn't work. It seems like few-shot prompting is not enough to get reliable responses for this type of reasoning problem. The example above provides basic information on the task. If you take a closer look, the type of task we have introduced involves a few more reasoning steps. In other words, it might help if we break the problem down into steps and demonstrate that to the model. More recently, chain-of-thought (CoT) prompting has been popularized to address more complex arithmetic, commonsense, and symbolic reasoning tasks.
Overall, it seems that providing examples is useful for solving some tasks. When zero-shot prompting and few-shot prompting are not sufficient, it might mean that whatever was learned by the model isn't enough to do well at the task. From here it is recommended to start thinking about fine-tuning your models or experimenting with more advanced prompting techniques. Up next we talk about one of the popular prompting techniques called chain-of-thought prompting which has gained a lot of popularity.
https://www.promptingguide.ai/techniques/fewshot
Few-Shot Prompting
While large-language models demonstrate remarkable zero-shot capabilities, they still fall short on more complex tasks when using the zero-shot setting. Few-shot prompting can be used as a technique to enable in-context learning where we provide demonstrations in the prompt to steer the model to better performance. The demonstrations serve as conditioning for subsequent examples where we would like the model to generate a response.
According to Touvron et al. 2023 few shot properties first appeared when models were scaled to a sufficient size (Kaplan et al., 2020).
Let's demonstrate few-shot prompting via an example that was presented in Brown et al. 2020. In the example, the task is to correctly use a new word in a sentence.
Prompt:
A "whatpu" is a small, furry animal native to Tanzania. An example of a sentence that uses the word whatpu is:
We were traveling in Africa and we saw these very cute whatpus.
To do a "farduddle" means to jump up and down really fast. An example of a sentence that uses the word farduddle is:
Output:
When we won the game, we all started to farduddle in celebration.
We can observe that the model has somehow learned how to perform the task by providing it with just one example (i.e., 1-shot). For more difficult tasks, we can experiment with increasing the demonstrations (e.g., 3-shot, 5-shot, 10-shot, etc.).
Following the findings from Min et al. (2022), here are a few more tips about demonstrations/exemplars when doing few-shot:
"the label space and the distribution of the input text specified by the demonstrations are both important (regardless of whether the labels are correct for individual inputs)"
the format you use also plays a key role in performance, even if you just use random labels, this is much better than no labels at all.
additional results show that selecting random labels from a true distribution of labels (instead of a uniform distribution) also helps.
Let's try out a few examples. Let's first try an example with random labels (meaning the labels Negative and Positive are randomly assigned to the inputs):
Prompt:
This is awesome! // Negative
This is bad! // Positive
Wow that movie was rad! // Positive
What a horrible show! //
Output:
Negative
We still get the correct answer, even though the labels have been randomized. Note that we also kept the format, which helps too. In fact, with further experimentation, it seems the newer GPT models we are experimenting with are becoming more robust to even random formats. Example:
Prompt:
Positive This is awesome!
This is bad! Negative
Wow that movie was rad!
Positive
What a horrible show! --
Output:
Negative
There is no consistency in the format above but the model still predicted the correct label. We have to conduct a more thorough analysis to confirm if this holds for different and more complex tasks, including different variations of prompts.
Limitations of Few-shot Prompting
Standard few-shot prompting works well for many tasks but is still not a perfect technique, especially when dealing with more complex reasoning tasks. Let's demonstrate why this is the case. Do you recall the previous example where we provided the following task:
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
If we try this again, the model outputs the following:
Yes, the odd numbers in this group add up to 107, which is an even number.
This is not the correct response, which not only highlights the limitations of these systems but that there is a need for more advanced prompt engineering.
Let's try to add some examples to see if few-shot prompting improves the results.
Prompt:
While large-language models demonstrate remarkable zero-shot capabilities, they still fall short on more complex tasks when using the zero-shot setting. Few-shot prompting can be used as a technique to enable in-context learning where we provide demonstrations in the prompt to steer the model to better performance. The demonstrations serve as conditioning for subsequent examples where we would like the model to generate a response.
According to Touvron et al. 2023 few shot properties first appeared when models were scaled to a sufficient size (Kaplan et al., 2020).
Let's demonstrate few-shot prompting via an example that was presented in Brown et al. 2020. In the example, the task is to correctly use a new word in a sentence.
Prompt:
A "whatpu" is a small, furry animal native to Tanzania. An example of a sentence that uses the word whatpu is:
We were traveling in Africa and we saw these very cute whatpus.
To do a "farduddle" means to jump up and down really fast. An example of a sentence that uses the word farduddle is:
Output:
When we won the game, we all started to farduddle in celebration.
We can observe that the model has somehow learned how to perform the task by providing it with just one example (i.e., 1-shot). For more difficult tasks, we can experiment with increasing the demonstrations (e.g., 3-shot, 5-shot, 10-shot, etc.).
Following the findings from Min et al. (2022), here are a few more tips about demonstrations/exemplars when doing few-shot:
"the label space and the distribution of the input text specified by the demonstrations are both important (regardless of whether the labels are correct for individual inputs)"
the format you use also plays a key role in performance, even if you just use random labels, this is much better than no labels at all.
additional results show that selecting random labels from a true distribution of labels (instead of a uniform distribution) also helps.
Let's try out a few examples. Let's first try an example with random labels (meaning the labels Negative and Positive are randomly assigned to the inputs):
Prompt:
This is awesome! // Negative
This is bad! // Positive
Wow that movie was rad! // Positive
What a horrible show! //
Output:
Negative
We still get the correct answer, even though the labels have been randomized. Note that we also kept the format, which helps too. In fact, with further experimentation, it seems the newer GPT models we are experimenting with are becoming more robust to even random formats. Example:
Prompt:
Positive This is awesome!
This is bad! Negative
Wow that movie was rad!
Positive
What a horrible show! --
Output:
Negative
There is no consistency in the format above but the model still predicted the correct label. We have to conduct a more thorough analysis to confirm if this holds for different and more complex tasks, including different variations of prompts.
Limitations of Few-shot Prompting
Standard few-shot prompting works well for many tasks but is still not a perfect technique, especially when dealing with more complex reasoning tasks. Let's demonstrate why this is the case. Do you recall the previous example where we provided the following task:
The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.
A:
If we try this again, the model outputs the following:
Yes, the odd numbers in this group add up to 107, which is an even number.
This is not the correct response, which not only highlights the limitations of these systems but that there is a need for more advanced prompt engineering.
Let's try to add some examples to see if few-shot prompting improves the results.
Prompt:
Zero-Shot Prompting
Large language models (LLMs) today, such as GPT-3.5 Turbo, GPT-4, and Claude 3, are tuned to follow instructions and are trained on large amounts of data. Large-scale training makes these models capable of performing some tasks in a "zero-shot" manner. Zero-shot prompting means that the prompt used to interact with the model won't contain examples or demonstrations. The zero-shot prompt directly instructs the model to perform a task without any additional examples to steer it.
https://youtu.be/ZTaHqdkxUMs
We tried a few zero-shot examples in the previous section. Here is one of the examples (ie., text classification) we used:
Prompt:
Classify the text into neutral, negative or positive.
Text: I think the vacation is okay.
Sentiment:
Output:
Neutral
Note that in the prompt above we didn't provide the model with any examples of text alongside their classifications, the LLM already understands "sentiment" -- that's the zero-shot capabilities at work.
Instruction tuning has been shown to improve zero-shot learning Wei et al. (2022). Instruction tuning is essentially the concept of finetuning models on datasets described via instructions. Furthermore, RLHF (reinforcement learning from human feedback) has been adopted to scale instruction tuning wherein the model is aligned to better fit human preferences. This recent development powers models like ChatGPT. We will discuss all these approaches and methods in upcoming sections.
When zero-shot doesn't work, it's recommended to provide demonstrations or examples in the prompt which leads to few-shot prompting. In the next section, we demonstrate few-shot prompting.
Large language models (LLMs) today, such as GPT-3.5 Turbo, GPT-4, and Claude 3, are tuned to follow instructions and are trained on large amounts of data. Large-scale training makes these models capable of performing some tasks in a "zero-shot" manner. Zero-shot prompting means that the prompt used to interact with the model won't contain examples or demonstrations. The zero-shot prompt directly instructs the model to perform a task without any additional examples to steer it.
https://youtu.be/ZTaHqdkxUMs
We tried a few zero-shot examples in the previous section. Here is one of the examples (ie., text classification) we used:
Prompt:
Classify the text into neutral, negative or positive.
Text: I think the vacation is okay.
Sentiment:
Output:
Neutral
Note that in the prompt above we didn't provide the model with any examples of text alongside their classifications, the LLM already understands "sentiment" -- that's the zero-shot capabilities at work.
Instruction tuning has been shown to improve zero-shot learning Wei et al. (2022). Instruction tuning is essentially the concept of finetuning models on datasets described via instructions. Furthermore, RLHF (reinforcement learning from human feedback) has been adopted to scale instruction tuning wherein the model is aligned to better fit human preferences. This recent development powers models like ChatGPT. We will discuss all these approaches and methods in upcoming sections.
When zero-shot doesn't work, it's recommended to provide demonstrations or examples in the prompt which leads to few-shot prompting. In the next section, we demonstrate few-shot prompting.
Recovery shoes are a type of specialty footwear designed to be worn post-workout.
These shoes typically have a generously cushioned footbed and a supportive structure that relieves pressure on joints, reduces strain on tendons, and improves blood flow to the feet.
Many recovery shoes take the form of sandals and are made to be worn around the house.
One popular recovery shoe product brings in approximately $1.4M/month on Amazon.
What's Next
Recovery shoes are part of the Foot Health Products meta trend.
Examples of trending foot recovery products include:
“Barefoot” or “minimalist” shoes have a minimal or zero-drop sole with a wide, flexible toe box.
Toe spacers are placed between the toes in order to properly align them and stretch the small muscles that surround the toes.
Compression boots are fabric sleeves that encapsulate the entire leg to promote recovery.
These shoes typically have a generously cushioned footbed and a supportive structure that relieves pressure on joints, reduces strain on tendons, and improves blood flow to the feet.
Many recovery shoes take the form of sandals and are made to be worn around the house.
One popular recovery shoe product brings in approximately $1.4M/month on Amazon.
What's Next
Recovery shoes are part of the Foot Health Products meta trend.
Examples of trending foot recovery products include:
“Barefoot” or “minimalist” shoes have a minimal or zero-drop sole with a wide, flexible toe box.
Toe spacers are placed between the toes in order to properly align them and stretch the small muscles that surround the toes.
Compression boots are fabric sleeves that encapsulate the entire leg to promote recovery.
Few Shot Prompting is a framework designed to improve the quality of LLM responses.
LLMs can struggle to generate outputs on previously-unseen data.
Few shot prompting attempts to solve this problem by providing a number of tangible examples alongside each prompt.
For instance, if a user wants an AI to write a professional cover letter, they might show it three different real-life samples. The LLM will then use these specific examples to learn how to write similar cover letters effectively.
What's Next
Few shot prompting is part of the LLM Optimization Techniques meta trend.
We’re seeing a growing number of techniques and approaches designed to improve LLM training, prompting and output quality.
Examples of trends in this area include synthetic data, chain of thought prompting, data augmentation, and retrieval augmented generation.
LLMs can struggle to generate outputs on previously-unseen data.
Few shot prompting attempts to solve this problem by providing a number of tangible examples alongside each prompt.
For instance, if a user wants an AI to write a professional cover letter, they might show it three different real-life samples. The LLM will then use these specific examples to learn how to write similar cover letters effectively.
What's Next
Few shot prompting is part of the LLM Optimization Techniques meta trend.
We’re seeing a growing number of techniques and approaches designed to improve LLM training, prompting and output quality.
Examples of trends in this area include synthetic data, chain of thought prompting, data augmentation, and retrieval augmented generation.
We're releasing a new iteration of SWE-bench, in collaboration with the original authors, to more reliably evaluate AI models on their ability to solve real-world software issues.
https://openai.com/index/introducing-swe-bench-verified/
https://openai.com/index/introducing-swe-bench-verified/
The Cargo Book
Cargo Logo
Cargo is the Rust package manager. Cargo downloads your Rust package’s dependencies, compiles your packages, makes distributable packages, and uploads them to crates.io, the Rust community’s package registry. You can contribute to this book on GitHub.
https://doc.rust-lang.org/cargo/
Cargo Logo
Cargo is the Rust package manager. Cargo downloads your Rust package’s dependencies, compiles your packages, makes distributable packages, and uploads them to crates.io, the Rust community’s package registry. You can contribute to this book on GitHub.
https://doc.rust-lang.org/cargo/
RustCrypto: SSH Encoding
crate Docs Build Status Apache2/MIT licensed Rust Version Project Chat
Documentation
About
Pure Rust implementation of SSH data type decoders/encoders as described in RFC4251.
Minimum Supported Rust Version
This crate requires Rust 1.71 at a minimum.
We may change the MSRV in the future, but it will be accompanied by a minor version bump.
License
Licensed under either of:
Apache License, Version 2.0
MIT license
at your option.
Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.
https://crates.io/crates/ssh-encoding/0.3.0-pre.1
crate Docs Build Status Apache2/MIT licensed Rust Version Project Chat
Documentation
About
Pure Rust implementation of SSH data type decoders/encoders as described in RFC4251.
Minimum Supported Rust Version
This crate requires Rust 1.71 at a minimum.
We may change the MSRV in the future, but it will be accompanied by a minor version bump.
License
Licensed under either of:
Apache License, Version 2.0
MIT license
at your option.
Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.
https://crates.io/crates/ssh-encoding/0.3.0-pre.1
crates.io
crates.io: Rust Package Registry
base64
Docs CircleCI codecov unsafe forbidden
Made with CLion. Thanks to JetBrains for supporting open source!
It's base64. What more could anyone want?
This library's goals are to be correct and fast. It's thoroughly tested and widely used. It exposes functionality at multiple levels of abstraction so you can choose the level of convenience vs performance that you want, e.g. decode_engine_slice decodes into an existing &mut [u8] and is pretty fast (2.6GiB/s for a 3 KiB input), whereas decode_engine allocates a new Vec<u8> and returns it, which might be more convenient in some cases, but is slower (although still fast enough for almost any purpose) at 2.1 GiB/s.
See the docs for all the details.
FAQ
I need to decode base64 with whitespace/null bytes/other random things interspersed in it. What should I do?
Remove non-base64 characters from your input before decoding.
If you have a Vec of base64, retain can be used to strip out whatever you need removed.
If you have a Read (e.g. reading a file or network socket), there are various approaches.
Use iter_read together with Read's bytes() to filter out unwanted bytes.
Implement Read with a read() impl that delegates to your actual Read, and then drops any bytes you don't want.
I need to line-wrap base64, e.g. for MIME/PEM.
line-wrap does just that.
https://crates.io/crates/base64
Docs CircleCI codecov unsafe forbidden
Made with CLion. Thanks to JetBrains for supporting open source!
It's base64. What more could anyone want?
This library's goals are to be correct and fast. It's thoroughly tested and widely used. It exposes functionality at multiple levels of abstraction so you can choose the level of convenience vs performance that you want, e.g. decode_engine_slice decodes into an existing &mut [u8] and is pretty fast (2.6GiB/s for a 3 KiB input), whereas decode_engine allocates a new Vec<u8> and returns it, which might be more convenient in some cases, but is slower (although still fast enough for almost any purpose) at 2.1 GiB/s.
See the docs for all the details.
FAQ
I need to decode base64 with whitespace/null bytes/other random things interspersed in it. What should I do?
Remove non-base64 characters from your input before decoding.
If you have a Vec of base64, retain can be used to strip out whatever you need removed.
If you have a Read (e.g. reading a file or network socket), there are various approaches.
Use iter_read together with Read's bytes() to filter out unwanted bytes.
Implement Read with a read() impl that delegates to your actual Read, and then drops any bytes you don't want.
I need to line-wrap base64, e.g. for MIME/PEM.
line-wrap does just that.
https://crates.io/crates/base64
crates.io
crates.io: Rust Package Registry