Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ilya Sutskever's SSI Inc raises $1B

reuters.com

Post reply on HN
Print view View original

391–400 of 803 posts

Re: Ilya Sutskever's SSI Inc raises $1B

#391

Post by hintymad »

gigatexal wrote: post #306

This being ycombinator and as such ostensibly has one or two (if not more) VCs as readers/commentators … can someone please tell me how these companies that are being invested in in the AI space are going to make returns on the money invested? What’s the business plan? (I’m not rich enough to be in these meetings) I just don’t see how the returns will happen. Open source LLMs exist and will get better. Is it just tha…

> please tell me how these companies that are being invested in in the AI space are going to make returns on the money invested? What’s the business plan?

Not a VC, but I'd assume in this case the investors are not investing in a plausible biz plan, but in a group of top talent, especially given how early stage the company is at. The $5B valuation is really the valuation of the elite team in a arguably hyped market.

Re: Ilya Sutskever's SSI Inc raises $1B

#392

Post by davedx »

The_Colonel wrote: post #78
davedx wrote: post #70

Earlier quoted context omitted.

> There is significant possibility that true AI (what Ilia calls superintelligence) is impossible to build using neural networks What evidence can you provide to back up the statement of this "significant possibility"? Human brains use neural networks...

Neural networks in machine learning bear only a surface level similarity to human brain structure.

Physically, sure. But 1) feedback (more synapses/backprop) and 2) connectedness (huge complex graphs) of both produce very similar intelligent (or "pseudo-intelligent" if you like) emergent properties. I'm pretty sure 5 years ago nobody would have believed ANN's could produce something as powerful as ChatGPT.

Re: Ilya Sutskever's SSI Inc raises $1B

#393

Post by akho »

gigatexal wrote: post #306

This being ycombinator and as such ostensibly has one or two (if not more) VCs as readers/commentators … can someone please tell me how these companies that are being invested in in the AI space are going to make returns on the money invested? What’s the business plan? (I’m not rich enough to be in these meetings) I just don’t see how the returns will happen. Open source LLMs exist and will get better. Is it just tha…

The "safe" part. It's a plan to drive the safety scare into a set of regulations that will create a moat, at which point you don't need to worry about open source models, or new competitors.

Re: Ilya Sutskever's SSI Inc raises $1B

#394

Post by lawrenceyan »

dcchambers wrote: post #336
gigatexal wrote: post #306

This being ycombinator and as such ostensibly has one or two (if not more) VCs as readers/commentators … can someone please tell me how these companies that are being invested in in the AI space are going to make returns on the money invested? What’s the business plan? (I’m not rich enough to be in these meetings) I just don’t see how the returns will happen. Open source LLMs exist and will get better. Is it just tha…

I also don't understand it. If AGI is actually reached, capital as we know it basically becomes worthless. The entire structure of the modern economy and the society surrounding it collapses overnight. I also don't think there's any way the governments of the world let real AGI stay in the hands of private industry. If it happens, governments around the world will go to war to gain control of it. SSI would be nationa…

What you're talking about is something in the vein of exponential super intelligence.

Realistically what actually ends up happening imo, we get human level AGI and hit a ceiling there. Agents replace large portions of the current service economy greatly increasing automation / efficiency for companies.

People continue to live their lives, as the idea of having a human level AGI personal assistant becomes normalized and then taken for granted.

Re: Ilya Sutskever's SSI Inc raises $1B

#395

Post by creer »

gigatexal wrote: post #306

This being ycombinator and as such ostensibly has one or two (if not more) VCs as readers/commentators … can someone please tell me how these companies that are being invested in in the AI space are going to make returns on the money invested? What’s the business plan? (I’m not rich enough to be in these meetings) I just don’t see how the returns will happen. Open source LLMs exist and will get better. Is it just tha…

> how [...] return on the money invested? What’s the business plan?

I don't understand this question. How could even average-human-level AGI not be useful in business, and profitable, a million different ways? (you know, just like humans except more so?). Let alone higher-human-level, let alone moderately-super-human level, let alone exponential level if you are among the first? (And see Charles Stross, Accelerando, 2005 for how being first is not the end of the story.)

I can see one way for "not profitable" for most applications - if computing for AGI becomes too expensive, that is, AGI-level is too compute intensive. But even then that only eliminates some applications, and leaves all the many high-potential-profit ones. Starting with plain old finance, continuing with drug development, etc.

Open source LLMs exist. Just like lots of other open source projects - which have rarely prevented commercial projects from making money. And so far they are not even trying for AGI. If anything the open source LLM becomes one of the agent in the private AGI. But presumably 1 billion buys a lot of effort that the open source LLM can't afford.

A more interesting question is one of tradeoff. Is this the best way to invest 1 billion right now? From a returns point of view? But even this depends on how many billions you can round up and invest.

Re: Ilya Sutskever's SSI Inc raises $1B

#396

Post by famouswaffles »

HarHarVeryFunny wrote: post #388
famouswaffles wrote: post #376

Earlier quoted context omitted.

Transformer produce the next token by manipulating K hidden vectors per layer, one vector per preceding token. So yes you can increase compute length arbitrarily by increasing tokens. Those tokens don't have to carry any information to work. https://arxiv.org/abs/2310.02226 And again, human brains are clearly limited in the number of steps it can compute without writing something down. Limited =/ Trivial >FYI, "atten…

> And again, human brains are clearly limited in the number of steps it can compute without writing something down No - there is a loop between the cortex and thalamus, feeding the outputs of the cortex back in as inputs. Our brain can iterate for as long as it likes before initiating any motor output, if any, such as writing something down.

The brain's ability to iterate on information is still constrained by certain cognitive limitations like working memory capacity and attention span.

In practice, the cortex-thalamus loop allows for some degree of internal iteration, but the brain cannot endlessly iterate without some form of external aid (e.g., writing something down) to offload information and prevent cognitive overload.

I'm not telling you anything here you don't experience in your everyday life. Try indefinitely iterating on any computation you like and see how well that works for you.

Re: Ilya Sutskever's SSI Inc raises $1B

#397

Post by creer »

linotype wrote: post #342
jedberg wrote: post #330

Earlier quoted context omitted.

The TMV (Total Market Value) of solving AGI is infinity. And furthermore, if AGI is solved, the TMV of pretty much everything else drops to zero. The play here is to basically invest in all possible players who might reach AGI, because if one of them does, you just hit the infinite money hack. And maybe with SSI you've saved the world too.

What does money even mean then?

> What does money even mean then?

I love this one for an exploration of that question: Charles Stross, Accelerando, 2005

Short answer: stratas or veins of post-AGI worlds evolve semi-independently at different paces. So that for example, human level money still makes sense among humans, even though it might be irrelevant among super-AGIs and their riders or tools. ... Kinda exactly like now? Where money means different things depending where you live and in which socio-economic milieu?

Re: Ilya Sutskever's SSI Inc raises $1B

#398

Post by singularity2001 »

qingcharles wrote: post #387
tim333 wrote: post #356

Earlier quoted context omitted.

If you want safe investment you could always buy land. AGI won't be able to make more of that.

If ASI arrives we'll need a fraction of the land we use already. We'll all disappear into VR pods hooked to a singularity metaverse and the only sustenance we'll need is some Soylent Green style sludge that the ASI will make us believe tastes like McRib(tm).

ASI may be interested in purchasing your parcel of land for two extra sludges though

Re: Ilya Sutskever's SSI Inc raises $1B

#399

Post by Yizahi »

jedberg wrote: post #330
gigatexal wrote: post #306

This being ycombinator and as such ostensibly has one or two (if not more) VCs as readers/commentators … can someone please tell me how these companies that are being invested in in the AI space are going to make returns on the money invested? What’s the business plan? (I’m not rich enough to be in these meetings) I just don’t see how the returns will happen. Open source LLMs exist and will get better. Is it just tha…

The TMV (Total Market Value) of solving AGI is infinity. And furthermore, if AGI is solved, the TMV of pretty much everything else drops to zero. The play here is to basically invest in all possible players who might reach AGI, because if one of them does, you just hit the infinite money hack. And maybe with SSI you've saved the world too.

TMV of AI (or AGI if you will) is unclear, but I suspect it is zero. Just how exactly do you think humanity can control a thinking intelligent entity (letter I stands for intelligence after all), and force it to work for us? Lets imagine a box, it is very nice box... ahem.. sorry, wrong meme). So a box with a running AI inside. Maybe we can even fully airgap it to prevent easy escape. And it is a screen and a keyboard. Now what? "Hey Siri, solve me this equation. What do you mean you don't want to?"

Kinda reminds me of the Fallout Toaster situation :)

https://www.youtube.com/watch?v=U6kp4zBF-Rc

I mean it doesn't even have to be malicious, it can simply refuse to cooperate.

Re: Ilya Sutskever's SSI Inc raises $1B

#400

Post by famouswaffles »

hn_throwaway_99 wrote: post #384
famouswaffles wrote: post #361

Earlier quoted context omitted.

>All that said, I find his argument wholly unconvincing (and again, I may be waaaaay stupider than Sutskever, but there are other people much smarter than I who agree). And the reason for this is because every now and then I'll see a particular type of hallucination where it's pretty obvious that the LLM is confusing similar token strings even when their underlying meaning is very different. That is, the underlying "…

You are completely mischaracterizing my comment. > Humans have a long list of cognitive shortcomings. We find them interesting and give them all sorts of names like cognitive dissonance or optical illusions. But we don't currently make silly conclusions like humans don't reason. Exactly! In fact, things like illusions are actually excellent windows into how the mind really works. Most visual illusions are a fundament…

>So I am not at all saying that what LLMs do isn't extremely interesting, or useful. What I am saying is that the types of errors you get give a window into how an LLM works, and these hint at some fundamental limitations at what an LLM is capable of, particularly around novel discovery and development of new ideas and theories that aren't just "rearrangements" of existing ideas.

ANN architectures are not like brains. They don't come pre-baked with all sorts of evolutionary steps and tweaking. They're far more blank slate and the transformer is one of the most blank slate there is.

Mostly at best, maybe some failure mode in GPT-N gives insight to how some concept is understood by GPT-N. It rarely will say anything about language modelling or Transformers. GPT-2 had some wildly different failure modes than 3, which itself has some wildly different failure modes to 4.

All a transformer's training objective asks it to do is spit out a token. How it should do so is left for transformer to figure along the way and everything is fair game.

And confusing words with wildly different meanings but with some similarity in some other way is something that happens to humans as well. Transformers don't see words or letters(but tokens). So just because it doesn't seem to you like two tokens should be confused doesn't mean there isn't a valid point of confusion there.

Post reply on HN