zlacker

[return to "Ilya Sutskever "at the center" of Altman firing?"]
1. andrew+l5[view] [source] 2023-11-18 03:20:39
>>apsec1+(OP)
If Ilya unknown-to-anyone thinks people prefer him to Altman, he has another thing coming. I'm not an Altman fanboy, but anyone can see Altman is a rockstar and richt or wrong, that matters HUGELY to OpenAI.

If it's truly about a power play then this will be undone pretty quick, along with the jobs of the people who made it happen.

Microsoft has put a vast fortune into this operation and if Satya doesn't like this then it will be changed back real fast, Ilya fired and the entire board resign. That's my prediction.

◧◩
2. dannyk+16[view] [source] 2023-11-18 03:24:46
>>andrew+l5
Completely agree. Sam is on the phone with Satya as we speak.

Alternative is Sam goes in house to MS who already have all the weights of GPT-4 and build again, but constrained by any existing charter.

◧◩◪
3. andrew+F6[view] [source] 2023-11-18 03:28:50
>>dannyk+16
The most likely outcome at this stage IMO is Sam will start a new thing with a huge equity stake and just do it again.
◧◩◪◨
4. dinobo+qb[view] [source] 2023-11-18 04:06:44
>>andrew+F6
Sam is not an engineer. He can't do it without Ilya or Ilya-lite. And there are like 4 of those in the world.
◧◩◪◨⬒
5. andrew+hc[view] [source] 2023-11-18 04:14:04
>>dinobo+qb
>>He can't do it without Ilya or Ilya-lite.

The wrongest thing I've read on HN for a long while.

The world has alot more smart people in it than you realise, and Sam's rockstar profile gives him direct access to them.

◧◩◪◨⬒⬓
6. dinobo+8f[view] [source] 2023-11-18 04:34:31
>>andrew+hc
How many people do you think are capable of pushing the state of the art in AI research?

There is a massive amount of tooling and infrastructure involved. You can't just get some Andrew Ng Coursera guy off the street and buy 50,000 H100s at your local Fry's electronics. I wouldn't be surprised if there aren't even enough GPUs in the world for Altman to start a competitor in a reasonable amount of time.

I stand by my number, there are like 4 people in the world capable of building OpenAI. That is, a quality deep learning organization that pushes the state of the art in AI and LLMs.

Maybe you can find ~1,000 people in the world who can build a cheap knock-off that gets you to GPT3 (pre instruct) performance after about two years. But even that is no trivial effort.

◧◩◪◨⬒⬓⬔
7. dannyk+Ei[view] [source] 2023-11-18 04:57:04
>>dinobo+8f
Did you see Greg Brockmann also quit? Who do you think has contributed more to OpenAI code base??
[go to top]