hah, Microsoft will be in control from here on out whether they have someone technically on the board or not. They did the embrace and extend now we're on to extinguish.
But maybe the beginning of the end of OpenAI.
A week ago I was saying that it's likely the leading AI company in 5 years time hasn't been founded yet.
After the news Friday and looking more at how Ilya sees the future of neural networks, I actually thought there's a decent chance OpenAI might correct course to continue to lead.
If it becomes too productized under a strengthened Altman, it's back on the list of companies building into their own obsolescence.
The right way to the future is adapting alignment to the increasing complexity of the model. It's not just about 'safety' but about performance and avoiding Goodhart's Law.
The way all major players, OpenAI included, are handling that step is by carrying forward the techniques that were appropriate for less complex models. Which is a huge step back from the approach reflected very early on pre-release for GPT-4's chat model. An approach that seemed to reflect Ilya's vision.
As long as OpenAI keeps fine tuning to try to meet the low hanging fruit product demand they've created and screwing up their pretrained model on measures that haven't become the industry target, they aren't going to be competitive against yet to exist companies that don't take such naive approaches with the fine tuning step. Right now they have an advantage in Ilya being ahead of the trend, but if Altman returning is at the cost of Ilya's influence, they are going to continue to dig their long term grave in the pursuit of short term success.
I bet that in 5 years, we won't be talking about AI at all. Whatever can be squeezed from current technologies, will be squeezed much sooner.
Real AI will eventually be based on totally different principles. Maybe by utilizing the natural intelligence of living organisms (e.g. bacteria).