zlacker

Is the logic here that training a base model isn’t as easy or even possible in the same way that OpenAI did in the past, and that what they have in a trained model is valuable in that even with all the code and experience it couldn’t be reproduced today with new restrictions?

>>shoele+(OP)
Yes, potentially.

The data has to come from somewhere, and all of the outlets that were used to train ChatGPT, stable diffusion, etc. have since been locked down. Any new company that Sam Altman makes in the AI space won't be competing just on merits of talent and product, they will also need to pay for and negotiate access to data.

I'd actually expect this to get far worse going forward, now that other organizations have an idea of how valuable their data is. It's also trivial to justify locking it down under the guise of protecting people, privacy, etc.

>>codere+L8
Couldn’t he just partner with Microsoft and start with the Bing corpus?

>>rblatz+8i
What if Microsoft, et al, signed exclusivity with OpenAI?

>>piuant+0j
By the sounds of it, they just DID.