1/ As a data practitioner, you should be aware of the Data Privacy and Data Quality implications of large language models.  At the same time, you can think about how to use these tools to make your work more efficient. 2/ Before using ChatGPT, you should develop an understanding of what data is used for training its models, and what tasks it cannot handle.  For example, the model's training data is recent as of 2021 — so it lacks recent information.

1/ So apparently OpenAI at one point trained and ran a model with sign-flipped reward due to a coding bug 🤦‍♀️

"One of our code refactors introduced a bug which flipped the sign of the reward. Flipping the reward would usually produce incoherent text," 2/ "but the same bug also flipped the sign of the KL penalty. The result was a model which optimized for negative sentiment while preserving natural language. Since our instructions told humans to give very low ratings to continuations with sexually explicit text,"

1/ Getty Images is currently suing Stability AI, claiming they illegally scraped images from Getty properties to train the Stable Diffusion text-to-image model. Getty has put forward several examples where their watermark is included in images generated by the AI tool. 2/ The lawsuit marks a significant escalation in legal battles between content creators and the tech companies behind generative AI. Its outcome will set an initial standard for dealing with tradeoffs between original content creators and artificially generated content.

Thoughts on FTX 🧵

1/ People who know crypto should have asked more questions about how FTX seemed to be fine when others were down after Terra-Luna.

Even sound business struggle in times of turmoil. 2/ I'm guilty of not asking these questions.

It seemed like a case of "too good to be true..." because it was.