I think the argument was that Taylor was bigger relative to her peers than OpenAI is. But I would still dispute that.
NVIDIA is not a peer of OpenAI–they sell the shovels. Or the guns, if you want to view it that way. Microsoft is not a peer–they bought a significant chunk of OpenAI so they could gain access to them. Google is big, but it’s not clear that they’re a peer, as the competitive products they’ve shown have been disappointing. Apple and Amazon are not peers–again they are big, but they’ve barely shown up to the AI game. Their size comes from non-AI stuff.
I’ll agree that Taylor Swift is an existential threat to humanity. Wait–that is what you said, right?
Yeah, the market isn’t valuing OpenAI that much. It is betting that there will be a lot of research into AI moving forwards.
You can see this by looking at PE ratios, aka Price Earnings ratio. That’s the ratio of value of the company on the stock market to the company’s profits over the past year. High PE ratios indicate optimism about profit growth.
The S&P 500 currently has a PE of 20.2. Microsoft has a PE of 36.6, quite a bit higher which makes sense for a tech company with a promising future.
What about other tech companies with promising futures who are not so involved in AI? Well Apple has a PE of 31.3, same ballpark if a little lower. But Amazon has a PE of 76.8, much higher than Microsoft. So I don’t think Wall Street is making a big bet on AI, not that they are the final authority on this or anything else.
What about Nvidia, which makes chips that AI training models use? They have a PE of 117.7: Wall Street is very optimistic about their prospects. So the market is betting heavily on lots of research/investment in AI moving forwards: it’s not yet betting that the end product of OpenAI will be any good, at least in a big way. I’m not surprised: prospects for OpenAI profits 3+ years out are nebulous and Wall Street is often accused of short-termism.
All that said, yes this is a big news story: AI is a great issue of the day.
Yes–in the background of all this, there is a cultural war: the e/accs vs. the EAs. Or, accelerationists vs. decelerationists, or boomers vs. doomers, or whatever you want to call them. The EAs have gotten a black eye from the whole SBF/FTX thing, but their doomerist attitude still holds a lot of sway–and that seemingly drove the board’s decision in the first place. Then, there are the accelerationists who think the doomer stuff is a load of crap, and that we should be going full blast on AI, if for no other reason than that China certainly isn’t going to slow down, so we might as well be first.
Even if you think both camps have their heads up their asses, people are making real-world, multi-billion-dollar decisions based on the positions. So it produces tremendous drama.
There is no slowing down of AI development. It is too easy to do now and there is too much value to be extracted for the world to,voluntarily slow down. All slowing down will do is cause you to lose the race.
I agree, but the OpenAI events demonstrate that at least some people think a slowdown can have a meaningful effect.
You need about 10k H100s to train a large model, not to mention some other things. Those cost about $0.5B. So there’s still some barrier to entry. But it won’t be long before GPT-4 equivalent models are available everywhere (not to mention the generations beyond that).
Likewise, the handwringing about using copyrighted content will have no effect. The big companies may decide to exclude some data sets, but the data has already been scraped privately. It’ll be used to train the open models regardless. And outside the US and EU, if necessary.
Open source models are now around GPT-3.5 quality, with much smaller numbers of parameters. You can run GPT 3.5 level inference on a Mackbook with an M2 with pretty good performance.
Now that we are learning exactly what we need to do to accelerate LLMs, Numerous companies are now working on next-gen AI processors.
Even if you need a hundred million worth of GPUs, that kind of financing is within range of thousands of people and organizations. And if no one else is doing it, it makes their investment even more valuable. So no, this genie’s not going back in the bottle.
It was well-reported that at one point in the OpenAI shenanigans, Microsoft very nearly aqui-hired Altman, expected to take the OpenAI engineers with him, giving Microsoft most of OpenAI’s key resources without even the inconvenience of presenting an offer to the board. Nadella’s move here was being hailed as the most consequential and lucrative decision by a tech CEO in living memory.
Of course that didn’t go through, but if it had, that would have given Microsoft an incredible leg up over everyone in the AI space. It would have been headed back toward a near-monopoly on emergent business-critical technology (one that Microsoft has staked a big part of its future on).
There are unique elements here that makes it an interesting story:
OpenAI is a non-profit with several subsidiaries including a capped for-profit company.
The non-profit created a golden egg laying goose and are in the process of going wide with their product.
The DNA of the company (including employees and board members) includes a healthy dose of concern about the very things they are building.
It isn’t a traditional fight about who controls the company and stands to rake in the most money. It is a fight about the very nature of the non-profit’s charter and whether to rake in money at all. Remember this is the company that announced they built GPT-2, but refused to release it to the public because it was too dangerous.
This article has an interesting graph of the relationships between the various subsidiaries:
It seems to me that the big challenge of Microsoft “aqui-hiring” Altman and all of the OpenAI staff is that all of their work product to date would still belong to OpenAI. Any future product that OpenAI could argue was derivative of that work would open Microsoft up to lawsuits, which is why I imagine the final resolution was to return Altman to the company.
Microsoft has more lawyers than God, and moreover owns 49% of OpenAI. Given that such a lawsuit would be easily defensible simply by documenting design choices, and that they have considerable power over this situation, I’m not sure that was a driver here.
I could be wrong on that, I’m not an expert on the law or this corporate structure. But my point here is that this story is big news in no small part because a tech giant nearly cornered the current AI market, in the process stealing a big part of the pie from Elon Musk et. al.
As usual with Musk, that is doubtful. He invested, but as usual with him, exaggeration is the norm.
In fact, while the source of much of OpenAI’s funding remains unclear, filings contain only around $15 million of donations that can be traced definitively back to Musk.
I can’t find a cite quickly, but I believe Microsoft has a cross-licensing deal with OpenAI IP. Microsoft has invested $11B so I’d expect them to have some agreement in place. It still would be extremely disruptive to productivity.
I’ve seen a few newspaper articles that reference Steve Jobs’ ouster and rehiring at Apple. It’s a good story to draw in readers (and viewers) so gets lots of breathless coverage.