Some people were using OpenAI's advanced API parameters apparently to derive scientific insights about GPT3.5. This will break in under a week. Thank you OpenAI for making our point so eloquently and efficiently for us
(first image: introduction of our CUI'23 paper https://doi.org/10.1145/3571884.3604316 ; second image: OpenAI announces it will break getting information about prompt/output probabilities for whatever reason)
#OpeningUpChatGPT #proprietary #reproducibility #openscience
Credit where credit is due: sometimes folks releasing LLM+RLHFs actually do admit there are legal complications
"While I/we are proud of FAIR's open values/history, on this particular one we are still trying to navigate. Using Clueweb may have been a mistake in this regard as that is not free.. in general the product+legal landscape is very difficult these days :( .. just getting the paper release itself approved took many weeks..."
(FAIR=Facebook AI Research)
IEEE Spectrum's Michael Nolan covers our work on #OpeningUpChatGPT: "An assessment of openness among ostensibly open-source LLM models finds that few live up to the claim" https://spectrum.ieee.org/openai-not-open
With some choice quotes from @andreasliesenfeld and me
Mark Dingemanse
