zlacker

Yep.

What people don't realize is that Microsoft doesn't own the data or models that OpenAI has today. Yeah, they can poach all the talent, but it still takes an enormous amount of effort to create the dataset and train the models the way OpenAI has done it.

Recreating what OpenAI has done over at Microsoft will be nothing short of a herculean effort and I can't see it materializing the way people think it will.

>>leetha+(OP)
Microsoft has full access to code and weights as part of their deal.

>>leetha+(OP)
Except MSFT does have access to the IP, and MSFT has access to an enormous trove of their own data across their office suite, Bing, etc. It could be a running start rather than a cold start. A fork of OpenAI inside an unapologetic for profit entity, without the shackles of the weird board structure.

>>jdminh+z
Even if they don't, the OpenAI staff already know 99 ways to not make a good GPT model and can therefore skip those experiments much faster than anyone else.

>>leetha+(OP)
Correct. This is all really bad for Microsoft and probably great for Google. Yet, judging by price changes right now, markets don’t seem to understand this.

>>jdminh+z
> Even if they don't, the OpenAI staff already know 99 ways to not make a good GPT model and can therefore skip those experiments much faster than anyone else.

This unequivocally .... knowing not how to waste a very expensive training run is a great lesson

>>jdminh+z
Source for your statement?

>>leetha+(OP)
This comment is factually incorrect. As part of the deal with OpenAI, Microsoft has access to all of the IP, model weights, etc.

>>belter+x2
https://www.wsj.com/articles/microsoft-and-openai-forge-awkw...

> Some researchers at Microsoft gripe about the restricted access to OpenAI’s technology. While a select few teams inside Microsoft get access to the model’s inner workings like its code base and model weights, the majority of the company’s teams don’t, said the people familiar with the matter.