OpenAI won the current state of the adoption war because they focused on commercial availability and being willing to lose money on every API call more than others. There is no groundbreaking tech behind it.
>Microsoft take the code, model weights, everything of ChatGPT 4, and just continue developing it and produce a new MS-ChatGPT based on it?
They don't need Sam to do that, or his employees. With the current state of LLMs they can hire a few dozen people with some experience and that wouldn't be the hard part.
In my mind the aggressive poaching OpenAI has been doing with Meta and Google is because they have nothing special. If they did they wouldn't need to grow to 800 headcount and still hiring. They're anticipating some company will have the next breakthrough, and they wanted every important brain on their side
Having the weights to another transformer based LLM is great, but its not going to change humanity. The only thing I saw this weekend was a bunch of entrepreneurs who's new startup is 100% dependent on the GPT-4 API freaking out. Thats not ground breaking
crushes? sarcasm? Free Vicuna is 90% as good as the 100 Billion Dollar company made one. This makes my point.
And between Vicuna and GPT4-Turbo There are the Claude Models and WizardML. This is as even playing field as i've ever seen for the tech world.
When a $100B company gets to 2000, and the rest stay behind, thats when we know the tech has changed. Thats what OpenAI was hiring for. Not for anything they already gave Microsoft access to
If Microsoft wanted to get to 1100, they could do that with 50 smart hires from across the industry and universities. They don't need Sams teams $1M salaries for that
That's not how Elo scores work. A 120 point elo difference is a 66% win rate, which means you win 2:1. That means GPT4-Turbo wins twice as much as Vicuna. And even that isn't a fair comparison, since Elo isn't linear: it gets more and more challenging to go up in Elo as you go further and further.
Well, I did say they are willing to burn money more than others - making their commercial availability better. Better yet ask yourself - will hollowed out Open AI and/or Satya be willing to lose money on every API call? Are you willing to pay the full cost of running GPT-4?
Google et al have gambled the answer for most API consumers is no
We know you have heavy biases but its impossible to deny that OpenAI is leading both the commercial and open source space. I don't think that will always be true but it is true now.
On top of that, they have done an amazing job executing on getting product out to users. No other company in this space is even remotely close to this.
> They don't need Sam to do that, or his employees. With the current state of LLMs they can hire a few dozen people with some experience and that wouldn't be the hard part.
If OpenAI leadership views it as a part of their charter to release their weights, research, etc. as open source, ten other companies would immediately catch up to Microsoft.
> Isn't it the completely the opposite of that, though?
No, its more complicated, which is why they were initially actually open, then became closed as their assessment of the circumstances changed; their own inability to act as gatekeeper, which would go away with a split with Microsoft where Microsoft had unrestricted rights and was not in practice coordinating with them, could well sufficiently change the circumstances to change their stance, without any inconsistency in reasoning (not that reasoning itself can legitimately evolve, too.)
In simple terms, it plausible that they might see the broad preference as:
good gatekeeper of top of the line AI > open models > bad gatekeeper of top of the line AI
and also see Microsoft, after the current kerfuffle, as a bad gatekeeper.
>Microsoft take the code, model weights, everything of ChatGPT 4, and just continue developing it and produce a new MS-ChatGPT based on it?
They don't need Sam to do that, or his employees. With the current state of LLMs they can hire a few dozen people with some experience and that wouldn't be the hard part.
In my mind the aggressive poaching OpenAI has been doing with Meta and Google is because they have nothing special. If they did they wouldn't need to grow to 800 headcount and still hiring. They're anticipating some company will have the next breakthrough, and they wanted every important brain on their side
Having the weights to another transformer based LLM is great, but its not going to change humanity. The only thing I saw this weekend was a bunch of entrepreneurs who's new startup is 100% dependent on the GPT-4 API freaking out. Thats not ground breaking