Can't MSFT just use GPT-4 weights and retrain from there lol. They have permanent rights anyway right.
they will, indirectly tho, training their new model using syntethic data from gpt-4
If OpenAI is so great and MSFT owns 49% of them why would MSFT need to do something like this?
Why not put all that compute towards fine tuning models on product specific data
Because 49% isn’t 100%. It’s pretty clear every big tech company wants to control their own LLM.
Did he just take Pi with him?
I expect he just took his brain. Anything else might be cause for lawsuits. I can imagine a distopic fictional future where part of the brain that stores the last few years work is erased. Similar to the tv show Severance?