Hacker News Reader

It looks dire but it seems too early for the nearly-always-wrong naysayer crowd to be right & take their victory lap.

Maybe not. But I always thought that, especially with their lobbying efforts and desire to project a sense of seriousness (even if unfounded) around those efforts, that it was very unlikely they would release a next-gen model before the US election.

I agree that the marketing/messaging, esp on social media, is borderline deranged, swinging between “we are basically The Foundation” and “pweese try my pwoduct i hope u love it (heart hands emoji)”

> it seems too early for the nearly-always-wrong naysayer crowd to be right & take their victory lap.

I agree it's too early to call for sure, but just to clarify: The naysayers are nearly always right. We just only remember the times they were wrong.

> The naysayers are nearly always right. We just only remember the times they were wrong.

Wait what?

Nearly every venture of any sort fails. The only times we remember the naysayers are on the rare occasions where they were wrong and the venture succeeded.

Ah okay, that makes sense.

I'll let parent elaborate more on the intent, but the way I interpreted it was : Saying that a startup will fail (i.e. being a naysayer) and being right about that is the most likely outcome due to the current "success" distribution (most businesses/startups fail).

Also the most memorable ones are when people were dismissive but ultimately wrong about the viability of the business (like the "dropbox" comment).

I think there's a deeper implication that the naysayers _about the subject of the hype_ are usually right, rather than simply about anyone trying to exploit the hype. Metaverse was going to be the next big thing. Naysayers (correctly) laughed. Nobody talks about metaverse now.

Because we mostly remember the things that are still around. Everyone knows about Charles Darwin but nobody knows about Erasmus Darwin, kind of thing.

The vast majority of startups fail. Most attempts at business will fail. It's just the nature of things.

But I think a not uncommon pattern with "too big to fail" startups is that they can change their definition of success or failure in order to claim victory. Or at least, in order for Sam to do so.

They might not reach the stated goal of AGI or even general profitability, but if Sam and some key investors manage to come out ahead at the end of their maneuvering, then I'm sure they'll claim victory (and that they changed the world and all that pomp).

Funny how their definition of changing the world amounts to: Fuck you, got mine.

> Funny how their definition of changing the world amounts to: Fuck you, got mine.

It's sad, but at this point I pretty much assume anyone out of SV who claims they're trying to change the world are either liars, incompetents, or both. That whole culture has just been on a tear of goodwill-burning.

If it wasn't this way, it would mean new things are more likely to succeed than fail.

No.

See the waves of investor hype that immediately preceded the AI hypewave: Metaverse and Blockchain. The naysayers were absolutely right.

Not to worry though, the geniuses at Meta and a16z are sure that AI will stick the landing, after they bet the farm on Metaverse and Blockchain respectively.

Naysayers in the British establishment thought the 13 colonies would come crawling back when their little experiment with democracy failed and they needed some proper aristocrats who knew what they were doing.

The space race had naysayers; the NYT printed the opinion that manned flight would take one to ten million years a week before the Wright Brothers; the iPod was lame with less space than a Nomad; SpaceX and Tesla were both dismissed when they were young; 3D printing likewise dismissed as if it was only plastic trinkets; and I've seen plenty of people here on HackerNews proclaiming that "AI will never ${x}" for various things some of which had already been accomplished by that point.

There's always a naysayer. You can use their mere existence as evidence for much, it has to be the specific arguments.

There's a difference between a naysayer and criticism. Using your iPod example, it was described as lame, in a way to dismiss it out of hand. But it did indeed have less space than a Nomad. That's valid, but ultimately unimportant, criticism.

People here are criticizing the business fundamentals of the openAI company. What are your takes on its finances? Or are you just on board, everything is great, do not look behind the curtain where we hide our five billion dollar losses?

Investment return over the last 30 years:

S&P 6x

Nasdaq 24x

Sticking your money in tech may have ups and downs but has not done badly. Also AI will be huge in a way blockchain and the metaverse were not. Also re blockchain if you'd bought bitcoin rather than some bs startup you'd have done well. Though OpenAI may be in the bs startup category... dunno.

I haven’t owned anything on blockchain in my life.

But I’m not so sure the naysayers are right about blockchain though. The ecosystem is still growing and apparently replacing cash in countries that don’t have a functional currency system.

Which is not the promise of blockchain's hype artists. "Better than using Venezuelan bolivars" is not a high bar.

I would say the strawberry/o1 hype was even worse than the GPT-5 hype

There were months worth of articles on how strawberry is considered almost dangerous internally its so smart. I know we only got -mini and -preview but...this doesn't feel like AGI

> There were months worth of articles on how strawberry is considered almost dangerous internally its so smart.

Like clockwork, every time they need to drum up excitement:

https://www.theverge.com/2019/11/7/20953040/openai-text-gene...

I was bashing my head into walls since 2017 or so when people were saying AI will eat the world and we have to worry about non-alignment and I felt insane realizing no one else even asked if it was manufactured hype. People in my life to this day are still falling for these tactics despite, to me, seeming bright regarding everything else.

To be clear, it is true that transformers did change things but the merchants are still over selling it and everyone else laps it up while not meta-thinking about it for even one second.

It may be hype, but there's plenty of solid logic behind the general case.

There's also a huge range of practical demonstrations of non-aligned, monomaniacal and not particularly smart intelligences, that literally eat humans: bacteria.

(Also lions and tigers and bears, if you want to insist that evolution doesn't itself count as intelligence).

It's like in the 1990s - (a) the internet will revolutionize commerce so (b) you should buy Webvan. One can be true without the other and (a) took a while. But a lot of people bought Webvan and similar back in the day.

To be fair, o1 is a major breakthrough in the field. If other AI labs can't crack scaling useful inference compute, OpenAI will maintain a big lead.

Isn't o1 just applying last year's Tree of Thoughts paper in production? Is there any reason to believe that the other companies will struggle to implement their own?

https://github.com/princeton-nlp/tree-of-thought-llm

I don't think it's tree of thoughts at all.

I think it's as they say: reinforcement learning applied to cause it to generate a relatively long 'reasoning trace' of some kind from which the answer is obtained through summarisation.

I think it's likely a cleverly simplified version of QuietSTaR, with no thought tokens, just one big generation to which the RL is applied.

The way I believe it's trained in practice is as follows: they have a bunch of examples, some at the edge of GPT-4s ability to answer, some beyond it, some that GPT-4 can answer if you're lucky with the randomness. Then they give it one of these prompts, generate a fairly long text, maybe 3x the length of the answer, and summarize that to produce the final answer. Then they use REINFORCE to reward the generated texts that increase the probability of the summary being correct.

Not be be nitpicky, but being the first to deploy recent academic research papers to production should count as a breakthrough IMHO.

There seems to be several components involved: tree of thoughts, MCTS, RL, high quality chain of thought data, possibly multiple models.

o1 seems like it’s basically 4o with some chain of thought bolted on. Personally, I don’t consider chain of thought a breakthrough, let alone a major one.

It's much more than CoT. I suspect it will take other labs some time to replicate.

> o1 is a major breakthrough

Is it? I feel like if you don't care about the cost it's pretty easily replicable on any other LLM, just with a lang-chain sort of approach

Then why hasn't it been done?

Someone gave the various models an IQ test and the previous ones scored 80-90 so a bit dim compared to humans, and o1 got 120, quite bright for a human. https://www.reddit.com/r/ClaudeAI/comments/1fhwfyl/openai_1o...

this may have consequences for how useful it is.

CoT can be _easily_ achieved using langgraph in a similar manner. There’s no “scaling of inference” it’s just prompting, all the way down.

It's not just CoT

Totally agree. It took me a full week before I realized that the Strawberry/o1 model was the mysterious Q* Sam Altman has been hyping up for almost a full year since the openai coup, which... is pretty underwhelming tbh. It's an impressive incremental advancement for sure! But it's really not the paradigm shifting gpt-5 worthy launch we were promised.

Personal opinion: I think this means we've probably exhausted all the low hanging fruit in LLM land. This was the last thing I was reserving judgement for. When the most hyped up big idea openai has rn is basically "we're just gonna have the model dump out a wall of semi-optimized chain of thought every time and not send it over the wire" we're officially out of big ideas. Like I mean it obviously works... but that's more or less what we've _been_ doing for years now! Barring a total rethinking of LLM architecture, I think all improvements going forward will be baby steps for a while, basically moving at the same pace we've been going since gpt-4 launched. I don't think this is the path to AGI in the near term, but there's still plenty of headroom for minor incremental change.

By analogy, i feel like gpt-4 was basically the same quantum leap we got with the iphone 4: all the basic functionality and peripherals were there by the time we got iphone 4 (multitasking, facetime, the app store, various sensors, etc.), and everything since then has just been minor improvements. The current iPhone 16 is obviously faster, bigger, thinner, and "better" than the 4, but for the most part it doesn't really do anything extra that the 4 wasn't already capable of at some level with the right app. Similarly, I think gpt-4 was pretty much "good enough". LLMs are about as they're gonna get for the next little while, though they might get a little cheaper, faster, and more "aligned" (however we wanna define that). They might get slightly less stupid, but i don't think they're gonna get a whole lot smarter any time soon. Whatever we see in the next few years is probably not going to be much better than using gpt-4 with the right prompt, tool use, RAG, etc. on top of it. We'll only see improvements at the margins.

chat has become a limiting factor.

its both too linear, and hard to revise. it's hard to undo parts of the conversation that poison it. it's too hard to save important bits that shouldn't be forgotten or drowned out. its not word processor like enough.

i envision the next generation of these products being multipane by default.

on the left I have a chat, in the center I have a whiteboard, and on the right I have a rendered document. throwing a clip of something onto the whiteboard makes it modifiable by chat. "store this, categorize it, summarize it, place it in the document."

whatever comes next needs to function more like onenote or obsidian. just to use an example from current events today, lets say I want to make a Parody Dossier on Walz, similar to todays Vance leak. I should be able to describe the project to chat. It builds a document structure. I tell it we are going to scrape all of the internets jokes on Walz at a bbq or other non-scandals. I should be able to quickly click through a table of contents, and "chat" with each paragraph. "This one needs fleshing out, this one needs summarization." As we scrape a reddit post, we want to incorporate not only the original post, but all the best comments. I should be a be able to "chat with the document editor" and put together a 200 page document in the amount of time it took me to write this post, just by describing what is and isnt working, and dragging and dropping.

chat, the simplicity of it, understanding of complex sentences, and multi sentence conversations was a UI paradigm leap. it went well past keyword search and the new command line of the internet. its a great first step, and a nice reset after a decade of interface stagnation, but now its ubiquity and simplicity, like the search box, is clouding peoples imagination and ability to dream up the next new interactive interface, which I expect to involve more mouse and visual relationships.

tldr: the llm is a component of the next generation interface, not the entire interface itself.

I think LLMs are more iPhone version one and missing a lot of "all the basic functionality"

Re "all the basic functionality" if you are talking intelligence then you are kind of talking the things humans can do and LLMs kind of do basic language but are not much good and common sense reasoning, spatial awareness and shape rotation, motor skills and running around, emotional intelligence, figuring what people are thinking, planning and probably some other stuff. It'll be iPhone 4 when those are covered.

speaking of which, try asking ChatGPT how many r's are in strawberry

The problem is that we're going through what happened in computer vision all over again. Convnets were getting bigger and bigger. New improvements came about in training efficiency, making medium models better, etc. Until... marginal improvements became more and more marginal.

LLMs are going through the same thing now. Better and better every iteration but increasingly we're starting to see a fast approaching wall on what they can really do with the current paradigm.

We are literally standing on the precipice of agency. After that, it's over. You see us approaching a wall, I see a cliff.

To stick with the metaphor, it's not super unclear if agency is within grasp or on the other side of an abyss. LLMs are definitely an improvement, but it's not at all clear if they can scale to human-level agency. If they reach that, it's even more unclear if they could ever reach superhuman levels given that all their training data is human-level.

And finally, we can see from normal human society that it is hardly ever the smartest humans who achieve the most or rise to the highest levels of power. There is no reason to believe that an AI with agency would an inherent "it's over" scenario

We kill millions of humans with agency every day on this planet. And none of them immediately die if we stop providing them with electrical power ... well I guess a small amount of them do. Anyway, we'll be fine. If the AI come, can they immediately stop us from growing food? How can an AI prevent my orchard from giving apples next year? Sure, it can mess up Facebook, but at this point that'd be a benefit.

I don't want to be a hater but according to the UN about 61 million humans die per year, so that only comes out to ~167k per day rather than millions. Most of those will die from old age too, rather than "being killed".

Your main point is true though, even superhuman AI would have a rough time in actual combat. It's just too dependent on electricity and datacenters with known locations to actually have a chance.

I’m sure the superintelligent AI will convince humans to transport its core while plugged into a potato battery. But honestly did the supply chain attack on Lebanese pagers last week teach you nothing? AIs should be great at that.

What? Planting C4 in pagers? Those pagers did not blow up on their own. Hundreds of people needed to put all that explosive in all those devices.

Our world is still analogue.

> We kill millions of humans with agency every day on this planet. And none of them immediately die if we stop providing them with electrical power ... well I guess a small amount of them do. Anyway, we'll be fine. If the AI come, can they immediately stop us from growing food? How can an AI prevent my orchard from giving apples next year? Sure, it can mess up Facebook, but at this point that'd be a benefit.

1. IMHO, genocidal apocalypse scenarios like you describe are the wrong way to think about the societal danger of AGI. I think a far more likely outcome, and nearly as disastrous, is AGI eliminating the dependency of capital on labor, leading to a global China Shock on steroids (e.g. unimaginable levels of inequality and wealth concentration, which no level of personal adaptability could overcome).

2. Even in your apocalypse scenario, I think you underestimate the damage that could be done. I don't have an orchard, so I know I'm fucked if urban life-support systems get messed up over large enough area that no aid is coming to my local area afterwards. And a genocidal AI that wasn't blindingly stupid would wait until it had control of enough robots to act in the real world, after which it could agent-orange your orchard.

1. A massive loss of jobs like that would lead to a better redistribution of wealth. I don't believe the US would adequately react, but I trust the rest of the world to protect people from the problem.

2. Which robots?? Until robots can, without any human input, make any sort of other robots, then there is nothing to fear from AGI. Right now they barely walk, and the really big boys can solder one joint real good. That's not enough.

> A massive loss of jobs like that would lead to a better redistribution of wealth...

How do you come to that conclusion?

I think we'll get massive inequality because it'll be far easier to develop new technology than change the core ideology of capitalist society through some secular process. If a "better redistribution of wealth" were in the cards, it probably should have happened already, and probably cannot happen after the 99% loose most of their economic power by being replaced by automation.

If the 1% are forward-thinking, they'll push a UBI scheme to keep the 99% from getting too disruptive before they become irrelevant.

How can you shut down something that can multiply into data centers in different countries around the world?

I'm not sure you realize this, but computers control everything. A dumb bug shut down windows computers around the world a month ago. A smart AI could potentially rewrite every piece of software and lock us out of everything.

Those medicines your friends and family need to stay alive? Yea the factories that produce them only work if you do what the AI says.

> There is no reason to believe...

What is happening right now is so obvious that people have been predicting it for over 60 years. It is ingrained in our culture. Everyone knows what happens when the AI becomes smarter than us.

If you 'see no reason' then you are looking through a microscope. Lift your head up and look around. Agency isn't black and white, it is a gradient. We already have agency to some degree, and it is improving fast.

This reminds of all the anti-Tesla "Tesla is dead/bankrupt" in 2017/18 takes from mostly clueless people (Gary Marcus is a certified idiot) who don't know anything about scaling a business, TAM of Knowledge Industry ($30T and growing).

Always, happy to take the other side of the bet of popular HN comments. (see META, TSLA)

"The fact that some geniuses were laughed at does not imply that all who are laughed at are geniuses. They laughed at Columbus, they laughed at Fulton, they laughed at the Wright brothers. But they also laughed at Bozo the Clown."

You're welcome to whichever side you want to take, but you're not right by the simple virtue of being contrary.

And not right by virtue of $STOCK_GO_UP unless your entire premise is $STOCK_GO_UP.

Of course, being contrarian and being right is important.

Happy to bet on Sam Altman and go short on Gary Marcus and popular HN in this case.

All that proves is that shitty businesses that already have scale take a long while to crumble. (See also: Xitter)

I don't know if OpenAI has a moat or if they will succeed long term against competitors, and certainly the valuation is high, but pointing to the lack of GPT-5 already is laughable. The speed of improvements - including GPT 4o1 preview and the new voice mode - and the business deals (Apple/Siri, whatever he will use the massive capital he raises for) are astounding. The idea that they have been resting on their laurels and have nothing else to show is just demonstrably untrue to date. Maybe the exec departures are a canary in the coal mine for the future, but I'm willing to at least give them a couple years until the last time they shipped what is magical to me (most recently, yesterday).

> The speed of improvements - including GPT 4o1 preview and the new voice mode - and the business deals (Apple/Siri, whatever he will use the massive capital he raises for) are astounding.

They're also rapidly replicable. I don't believe for a moment that Apple is designing their system in a way that doesn't allow switching at a moment's notice, and everything they're doing is copied within 6 months by competitors including LLaMA, which Meta keeps releasing for free.

> The idea that they have been resting on their laurels and have nothing else to show is just demonstrably untrue to date.

I didn't assert that they were resting on their laurels, I asserted that they have no path forward to the next generation and AGI. If they did have a path they wouldn't keep burning their hype for it on applications of the previous generation that are easy to replicate.

What they were lobbying for until very recently isn't a moat, if anything it's the opposite: "we know anything less capable is fine, focus your attention on us and make sure what we do is safe".

What's changed very recently, which is a moat and which they may or may not get, is seven 5 GW data centres — the equivalent of "can we tile half of Rhode Island in PV?"

> we know anything less capable is fine, focus your attention on us and make sure what we do is safe

This is a form of regulatory capture. You get a lead against your competition and then persuade governments to make it hard for people to follow you by imposing rules and regulations.

Regulatory capture is used in this way to build moats.

> This is a form of regulatory capture. You get a lead against your competition and then persuade governments to make it hard for people to follow you by imposing rules and regulations.

They were not making it hard to follow them. That's the point. They were saying they themselves don't really know what they're doing, while saying open source should be protected from any regulations imposed on them.

Hard to go past them, perhaps; but even then, given the scale of compute for training bigger models, the only people who could even try would have an easy time following whatever rules are created.

> o1 is impressive

I agree, the capabilities and reasoning are really nice compared to 4o. But my first thought when testing it- how are they going to pay for it.. trivial questions burn 15s and more complex 30-40s of some processing time. How does this scale to millions of users??

What's with his obsession with GPT-5? Altman has conistently been saying that there will be no GPT-5 this year almost since the years beginning. He's acting like OpenAI promised GPT-5 and are unable to release it or something.

Will there even be a GTP-5?

Maybe next up is o2 then o3 and so on.