This actually seems like a decent compromise. Sam and Greg can retain velocity on the product side without having to spin up a whole new operation in direct competition with their old levers of power, and Ilya + co can remain in possession of the keys to the kingdom.
Maybe I'm reading too much into it, but for me it is us framed as if they won't be working on GPT-based products, but on research.
The whole thing reads like this to me: "In hindsight, we should've done more due diligence before developing a hard dependency on an organization and its product. We are aware that this was a mistake. To combat this, we will do damage control and continue to work with OpenAI, while developing our in-house solution and ditching this hard dependency. Sam & Co. will reproduce this and it will be fully under our control. So rest assured dear investors."
How do you conduct research with sales people? even if they manage to bring in researchers from OpenAI, the only gain here is microsoft getting some of the researchers behind the products and/or product developers.
Well, the same way a man with drive, discipline and money but very little in the way of technical expertise can build a company.
Sometimes you need someone who can drive a project and recruit the right people for the project. That person does not always need to be a subject matter expert.
Except they only had AI model velocity and not product velocity. The user-side implementation of chatGPT is actually quite below what would be expected based on their AI superiority. So the parts that Sam & Greg should be responsible for are actually not great.
Sam and Greg were responsible for everything including building the company, deciding on strategy, raising funding, hiring most of the team, coordinating the research, building the partnership with Microsoft and acquiring the huge array of enterprise customers.
To act like they were just responsible for the "UI parts" is ridiculous.
I'm the first to defend CEOs and it's not a popular position to be in usually, believe me. But in this case, they did an experiment and it blew up based on their model's superiority alone.
Product-wise, however, it's looking like good enough AI is being commoditized at the pace of weeks and days. They will be forced to compete on user experience and distribution vs the likes of Meta. So far OpenAI only managed to deliver additions that sound good on the surface but prove not to be sticky when the dust settles.
They have also been very dishonest. I remember Sam Altman said he was surprised no one built something like chat GPT before them. Well... people tried but 3rd parties were always playing catch-up because the APIs were waitlisted, censored, and nerfed.
a) Meta is not competing with OpenAI nor has any plans to.
b) AI is only being commoditised at the low-end for models that can be trained by ordinary people. At the high-end there is only companies like Microsoft, Google etc that can compete. And Sam was brilliant enough to lock in Microsoft early.
c) What was stopping 3rd parties from building a ChatGPT was the out of reach training costs not access to APIs which didn't even exist at the time.
a) Meta is training and releasing cutting-edge LLM models. When they manage to get the costs down, everyone and their grandma is going to have Meta's AI on their phone either through Facebook, Instagram, or Whatsapp.
b) Commoditization is actually mostly happening because companies (not individuals) are training the models. But that's also enough for commoditization to occur over time, even on higher-end models. If we get into the superintelligence territory, it doesn't even matter though, the world will be much different.
c) APIs for GPT were first teased as early as 2020s with broader access in 2021. They got implemented into 3rd party products but the developer experience of getting access was quite hostile early on. Chat-like APIs only became available after they were featured in ChatGPT. So Sam feigning surprise about others not creating something like it sooner with their APIs is not honest.
If I recall correctly, Mira Murati was actually the person responsible for productizing GPT into a Chatbot. Prior to that, OpenAI's plan was just to build models and sell API access until they reach AGI.
I know there's a lot of talk about Ilya, but if Sam poaches Mira (which seems likely at this point), I think OpenAI will struggle to build things people actually want, and will go back to being an R&D lab.
This is kind of true, I think programming even codellama or gpt3.5 is more than enough and gpt-4 is very nice but what is missing is good developer experience, and copy-pasting to the chat window is not that.
Ilya and co are going to get orphaned, there’s no point to the talent they have if they intend to slow things down so it’s not like they’ll remain competitive. The capacity that MSFT was going to sell to OpenAI will go to the internal team.
Maybe they want it that way and want to move on from all the LLM hype that was distracting them from their main charter of pushing the boundaries of AI research? If yes, then they succeeded handsomely
Articles like this on the internet are usually completely garbage, but Natalie is a really fantastic journalist. If you’re worried about woo because of the title, there isn’t any here (and quanta magazine is generally great for coverage of science)
tldr: there’s a calculation you can perform in quantum mechanics that suggests particles can tunnel through barriers faster than they could have traveled through free space at the speed of light. Scientists have now measured this more precisely, and this effect seems to hold up, but is still very small, so it’s hard to understand the ramifications of this for things like causality.
Right, articles like this are often short on crucial details, it's speculation until we see the research papers.
Re causality etc., as I mentioned above, if the duration of wavefunction collapse can actually be measured then this puts a whole new spin on the matter (sorry, puns aren't my forte but it seemed appropriate here). ;-)
Update, FYI, I've just found some links about the matter and posted them above. These describe an experiment where an electron was seen transiting smoothly between two energy states whilst being in two places at once.
Rationalists seem extremely common amongst the $300k+/year software engineers I know. If that isn’t “successful”, then I’m not sure we’re using the same word anymore.
You can argue that the causal arrow doesn’t point in the right direction—that is, that people that are successful just happen to wear the affectation of “rationalist” because that happens to be fashionable, and not that their rationalism led them to a “successful” career, but that doesn’t seem to hold up against scrutiny, in my experience. Of those that I personally know, most have been engaged with the community since at least the golden days of lesswrong.
> Rationalists seem extremely common amongst the $300k+/year software engineers I know
I'm a programmer who lives on the other side of the world (so no chance for me to make $300k+/year) and I must say that all this "rationalism" discussion makes me a little confused: do people really believe in this sort of stuff? Do they actually equate "success" with (mostly owning) "money"? Do they really think a "rational"(-ist) person would mainly think about how to earn (supposedly more) money? Why on Earth would he/she do that? Money is just a tool. Holding an important position in society (CEO, founder, whatever) is just a hindrance, it keeps one away from actually thinking about the stuff that really matters.
I'm pretty sure all this stuff was explained a lot better a long time ago by people a lot more smarter than me (right now I'm thinking at one of Plato's works, maybe "Symposium"? I'm not sure, I've last read many of them ~20 years ago), point is this specific "view of the world" seems very US-specific to me.
Earning lots of money sets you up to donate lots of money to effective charities. Many rationalists believe that's the best thing you can be doing for global quality-of-life.
If one understands that the value of your work is almost certainly higher than your salary, then you'd conclude to benefit the world the most, you'd be better allocating your work directly to the cause of global quality-of-life.
> Holding an important position in society (CEO, founder, whatever) is just a hindrance, it keeps one away from actually thinking about the stuff that really matters.
You may believe stuff like this, but many others don't, so it should be phrased as your opinion rather than as a fact.
Elon Musk, for example, has very strong beliefs about humanity's future on other planets. His position as founder/CEO of SpaceX lets him actually work towards making that dream a reality. You or I can dream all we want but we can't make them reality. If this is something you care about, then he's clearly successful in ways that we aren't, that are directly attributable solely to his role as wealthy person/founder.
I definitely recommend reading the sequences. More than anything, it's about recognizing cognitive biases. Altruism is where money comes in, but I would classify that as the Effective Altruism movement.
I think this is exactly right. Being successful in the top 0.1% is not actually rational. Even if you're brilliant, your chance of succeeding at that level is low. A truly rational person seeks out the best risk-adjusted return, not the best absolute return. If you're a technical minded person, making 300k/year in a software job is about the best risk-adjusted return of any profession I can think of.
Well yes, but many other countries have similar math when it comes to software engineering salaries as opposed to other fields. In most places, going into software is a good thing to do moneywise.
I find it difficult to find reliable numbers (since I work in a different industry), but it seems median entry salary for IT graduate is 40k€, which is 25k€ nett anually. I also find the middle-upper class difficult for germany (since salary upper class doesnt mean you are asset upperclass aka rich).
Personally I would say 25k€ nett is middle class; but here everybody wants to belong to the middle class so it usually has a wide stretch.
Big companies are unionized so salarys do not get that big, but are considered higher then in non-unionized companies.
What is the optimal risk adjustment ratio? And it is chosen by reason or by personal preference? Many software engineers are a bit risk averse and just do what they are told to do. Many entrepreneurs would rather experience the rushes of a business roller-coaster and walk away with no money than just sit and program as told.
There is no optimal ratio per se. You just divide gain by risk and maximize that quantity. This is a scale invariant function, at least, with respect to its maximum.
I think you must be omitting variables, or simply trying to maximize dollars or some other approximately ordered quantity rather than maximize subjective fulfillment in life.
Which variables are being omitted? Certainly there are other variables relevant to a quality life than money. But the rules apply to any utility function you might come up with. You want to maximize risk-adjusted return.
I’ve asked a number of people I know who are not “tech adjacent” (meaning that they don’t frequent boards like HN or tech Twitter, and they don’t live in the Bay Area) if they’ve ever heard of the Internet-based phenomenon that calls itself “rationalism”, and I have yet to find one who has heard of it. So I am skeptical about your claim regarding the causal arrow: I only know of this phenomenon because I’m in tech, and other tech people pointed me to it.
To me, this feels a little like saying “lots of tech people I know with high salaries were early contributors to Wikipedia, therefore being an early contributor to Wikipedia probably made them successful.”
I had become a rationalist(ish) at least a decade before I knew what rationalism was as a movement. I very similarly has lost my faith in religion, far before I ever knew what atheism as a movement was too.
It can be considered causal. In that, many people follow the thought processes that underlie rationalism which direct them towards careers and outcomes of the sort that the person above was mentioning. (95+ percentile salary as a stable independent contributor in a field that values logic and structured thought). They may do it while being completely ignorant of the movement, but still being a rationalist for all intents and purposes.
> They may do it while being completely ignorant of the movement, but still being a rationalist for all intents and purposes.
If you consider “rationalism” (in the sense of the subject of this thread) to be equivalent to “following a scientific or rational thought process”, then the main question asked in the article becomes nonsensical. “Where are all the successful people who followed rational thought processes?” is a genuinely foolish question, because you can find countless notable examples with no effort at all.
But of course that’s not what the post was asking, and it’s why the poster has a harder time answering the question. The post refers to the very specific Internet phenomenon of “rationalism” which, while cleverly incorporating the notion of rational thought into its name, actually refers to a specific group of people who follow a specific set of teachings.
And those people are massively concentrated in US tech and tech-adjacent areas, largely because that’s where this specific set of beliefs took off first. That’s the causal arrow here.
The number of people “qualified” to go to Harvard is probably more than 10x the current capacity of Harvard, though. Maintaining scarcity/exclusivity in the brand is more valuable than servicing the number of people technically qualified to go there, from Harvard’s perspective.
Yes, but many of those people choose to go to other prestigious schools. If Harvard admitted 50% of applicants, I don't see how you can claim quality wouldn't suffer.
You said that they should admit 10x. Presumably that would mean 50% rather than 5%. I think quality would go down, you said it wouldn't. Seems clear cut to me.
To elaborate a bit: people like Marcus tend to overload/move the goal posts with what the word “understand” means. I kinda feel like in a world where we have perfectly conversational chat bots that are capable of AI complete tasks—-that if these bots look like Chinese rooms under the hood, he’ll still be complaining that they don’t “understand” anything.
I don’t think it’s unreasonable to say that if you think something that doesn’t “understand” anything can do what GPT-2 can do, then maybe your definition of “understand” doesn’t cut reality at the joints
Understanding is not hard to understand. To understand is to reason from a model. Reasoning from a model is easy. Discovering the correct model is hard, analogous to the way that algebraic rules are easy, but finding the right equation for a particular problem is hard. Data trained NNs have neither a model, nor do they reason. QED
You could say that a trained neural net contains a model of how language works, and it reasons about sentences based on this model.
I think people are really hung up on that it has trouble reasoning about what its sentences are reasoning about, and skipping how amazing it is at reasoning about sentence structure itself.
Yes, but people don’t reason about language, they just do it. I know you think I’m confused about this but I’m not. I mean reason here quite explicitly because what we’re talking about is understanding. No one thinks that they ... uh, well ... “understand” language ... okay, we need a new word here because “understand” has two different meanings here. Let’s use “perform” for when you make correct choices from an inexplicit model, that’s what the NN does, and hold “understand” for what a linguist (maybe) does per language, and what a physicist does per orbital mechanics. What we are hoping a GAI will do is the latter. Any old animal can perform. Only humans, as far as we know, and perhaps a few others in relatively barrow cases, understand in the sense that a physicist understands OM. No NN trained on language is gonna have the present argument. Ever.
The subtlety here is that NNs do have a model, but it’s hard to see. Not just any neural network can perform as well as GPT-2–a very specific architecture can. That architecture, coupled with the data it’s trained on, implicitly represents a model, but it’s wildly obscured by the details of the architecture.
In this sense, people like Sutskever think that GPT-2 is a step on the path towards discovering the “correct” model.
It’s probably difficult to make much more progress without making extremely crisp by what you mean a “model” is, though, because I feel like it’s just as easy to move goal posts about what it means to “understand” as it does to “model”.
For example, replace every instance of “a model” in your post with “an understanding”, and it parses almost identically
I don’t understand your last point, but the point about it being hard to be clear about what a model means is exactly right. But it’s not because it’s not clear what a model is, but rather because it’s not clear what the modeling language of thought is. Here’s where the algebra analogy breaks down. Pretty obviously, the model or models that we are reasoning with in this discussion aren’t simple algebraic equations, but some sort of rich representations of cognitive science and computer science concepts. And, sure, there are NNs running those models, and NNs running the reasoning over them, but they have almost nothing to do with language in the sense of the syntax of sentences. Also, we didn’t get trained with eleventy zillion examples of AI discussions in order to form the models we are employing at this very moment.
The "programmable bacteria" concept seems a bit more general than the typical treatments that work in a petri dish / mice before flopping on humans. At the very least it's more novel.
