Here's an example from a paper that came out last month – so it wasn't trained on it.
@jeremyphoward
-
GPT-4 Recognizes Famous ML Equations Without Explicit Labels
By
–
I should note that these example equations are pretty famous – they're for cross-entropy and adam optimiser; obviously GPT4 recognised them, since it named the functions correctly, although I never said what they were. Perhaps for less-known equations it would be less good.
-
ChatGPT Completes Tests in Record Time
By
–
Cool in the time it took to write that ChatGPT finished running its tests
-
Custom instruction for concise responses using ‘vv’ notation
By
–
(Oh BTW if you're wondering about the 'vv' thing – it's a custom instruction I use when I want it to be concise)
-
GPT-4V for Mathematical Equation Understanding and Explanation
By
–
If you're like me and find it easier to read math than code, and you have access to @OpenAI GPT 4V, try pasting a image of an equation you wanna understand in there. It might just blow your mind.
-
Do Mistral Fine-Tunes Achieve Exceptional TruthfulQA Scores?
By
–
Do all mistral fine-tunes have insane truthfulqa scores?
-
Why MMLU Performance Lags Behind Other AI Evaluation Benchmarks
By
–
Very cool. I wonder why MMLU is much worse than other evals here
-
PCI Bandwidth Calculation for AI Hardware
By
–
Where do you get that PCI bandwidth calculation from? It's higher than I expected
