Chain-of-thought allows GPT-3 to solve problems it otherwise cannot solve at all. Consensus allows it to more reliably solve problems it can only solve some of the time. Here, we use both to demonstrate multi-step reasoning and recitation of factual knowledge.
@goodside
-
AI Reasoning Technique ‘Let’s Think Step by Step’ Discussed with Research Link
By
–
Next we apply "Let's think step by step," the method developed by Kojima et al. for improving accuracy on deductive reasoning problems. This encourages the model to lay out its reasoning, making it more likely to be right: https://
arxiv.org/abs/2205.11916 -
Using GPT-3 for a Step-by-Step MD5 Hash Calculation Task
By
–
In the Python script generated by the first instruction template, we give GPT-3 the following prompt: Q: What is the final character of the MD5 hash of the last digit of the release year of the Grimes album "Visions"?
A: Let's think step by step. -
Instruction Template Prompting Method for GPT-3 Code Generation
By
–
A simple demo of my "instruction template" method of prompting GPT-3, generating a Python 3 module using placeholders and notes within a template that contains only a single line of actual Python:
-
Implementing a Python Guess Function using GPT-3
By
–
Using GPT-3 to implement a `guess()` function in Python that returns whatever string seems reasonable for the context in which the function was called.
-
Complex GPT-3 Prompting Technique for MD5 Hash Determination
By
–
A GPT-3 prompt in instruction-templated Python yielding a valid Python completion that prompts GPT-3 again, using zero-shot chain-of-thought consensus, to determine the final character of the MD5 hash of the final digit of the release year of the album "Visions" by @Grimezsz
. -
Upcoming ChatGPT Features Found in Client Code: Pause, Copy, Prompt Palette
By
–
References to possible upcoming ChatGPT features found by @eeeziii in minified client JS:
– Pause completions
– Copy thread to clipboard
– “Prompt palette” / “Snippet palette” -
GPT-3 Synthesis of arXiv Preprint with BERT-embedded Hopfield Networks on Ethereum Blockchain
By
–
GPT-3 zero-shot synthesis of an arXiv pre-print in LaTeX notation on distributed manifold in-painting using BERT-embedded Hopfield networks on the Ethereum blockchain. Model text-davinci-003, temperature 0.7. Playground link: https://
beta.openai.com/playground/p/m
z75vuUjGsEusUyPheyEdcSZ?model=text-davinci-003
… -
Limitations of Generative AI Tuning and Data Distribution
By
–
Being able to produce outputs for an arbitrary input doesn’t give you the ability to reproduce the input distribution of the tuning data. Even if you somehow can, tuning on the generations of a tuned model is like making a copy of a copy of a VHS tape.
-
AI Model Behavior with Inappropriate Prompts and Content Filtering
By
–
I’m pretty sure that happens for inappropriate content at least. If you construct a prompt that looks benign in isolation but produces offensive output it spins for an unusually long time before saying “Sorry, can’t do that.”