[HN Gopher] The era of boundary-breaking advancements is over? [...
___________________________________________________________________
The era of boundary-breaking advancements is over? [video]
Author : randomgermanguy
Score : 52 points
Date : 2025-08-09 12:55 UTC (10 hours ago)
(HTM) web link (www.youtube.com)
(TXT) w3m dump (www.youtube.com)
| bpodgursky wrote:
| It is critical to remember that there is a market for people who
| say "AGI is not coming"
|
| It doesn't matter whether they are lying. People want to hear it.
| It's comforting. So the market fills the void, and people get
| views and money for saying it.
|
| Don't use the fact that people are saying it, as evidence that it
| is true.
| righthand wrote:
| Waiting for Agi-dot...
|
| The inverse can be true too: Just because people ARE saying
| that Agi is coming, isn't evidence that it is true.
| bpodgursky wrote:
| OK, but your null hypothesis should always be a first or
| second degree linear projection.
|
| "AI is getting better rapidly" is the current state of
| affairs. Arguing "AI is about to stop getting better" is the
| argument that requires strong evidence.
| camillomiller wrote:
| Compare Altman outlandish claims about GPT-5 and the
| reality of this update. Do you think they square out in any
| reasonable way?
| bpodgursky wrote:
| Please, please seriously think back to your 2020 self,
| and think about whether your 2020 self would be surprised
| by what AI can do today.
|
| You've frog-boiled yourself into timelines where "No
| WORLD SHAKING AI launches in the past 4 months" means "AI
| is frozen". In 4 months, you will be shocked if AI
| doesn't have a major improvement every 2 months. In 6
| months, you will be shocked if it doesn't have a major
| update ever 1 month.
|
| It's hard to see exponential curves while you're on it,
| I'm not trying to fault you here. But it's really
| important to stretch yourself to try.
| backpackviolet wrote:
| I'm still surprised by what AI can do. It's amazing. ...
| but I still have to double check when it's important that
| I get the right answer, I still have to review the code
| it writes, and I still am not sure there is actually
| enough business to cover what it will actually cost to
| run when it needs to pay for itself.
| th0ma5 wrote:
| To be honest, I had the right idea back then... This
| technology has fundamental qualities that require it to
| provide inaccurate token predictions that are only
| statistically probable. They aren't even trying to change
| this situation other than trying to find more data to
| train, saying you have to keep adding layers of them, or
| are saying it is the user's responsibility.
|
| There's been the obvious notion that digitizing the
| world's information is not enough and that hasn't
| changed.
| righthand wrote:
| What if I've not been impressed by giving a bunch of
| people a spam bot tuned to education materials? Am I frog
| boiled? Who cares about the actual advancement of this
| singular component if I was never impressed.
|
| You assume everyone is "impressed".
| jononor wrote:
| I for one am quite surprised. Sometimes impressed. But
| also often frustrated. And occasionally disappointed.
| Sometimes worried about the negative follow-on effects.
| Working with current LLMs spans the whole gamut... But
| for coding we are at the point where even the current
| level is quite useful. And as the tools/systems get
| better, the usefilness is going to increase quite a bit.
| Even if models improve slowly from this point on. It will
| impact the whole industry over the next years, and since
| software is eating the world, will impact many other
| industries as well. Exponential? Perhaps in the same way
| as computers and Internet have been exponential - cost
| per X (say tokens) will probably go down exponentially
| the next years and decades, the same way cost per FLOP
| went down, on megabytes transferred. But those
| exponential gains did not results in exponential growth
| in productivity, or if so, the exponent is much much
| lower. And I suspect it will likely be the same for
| artificial intelligence.
| backpackviolet wrote:
| > "AI is getting better rapidly"
|
| ... is it? I hear people saying that. I see "improvement":
| the art generally has the right number of fingers more
| often, the text looks like text, the code agents don't
| write stuff that even the linter says is wrong.
|
| But I still see the wrong number of fingers sometimes. I
| still see the chat bots count the wrong number of letters
| in a word. I still see agents invent libraries that don't
| exist.
|
| I don't know what "rapid" is supposed to mean here. It
| feels like Achilles and the Tortoise and also has the
| energy costs of a nation-state.
| righthand wrote:
| Agreed there really isn't any metrics that indicate this
| is true. Considering many models are still too complex to
| run locally. Llms are getting better for the corporations
| that sell access to them. Not necessarily for the people
| that use them.
| righthand wrote:
| "AI is getting better rapidly" is a false premise. As AI is
| a large domain. There is no way to quanitify the better as
| compared to the entire domain. "Llms are improving rapidly
| during a short period of time where they gain popularity"
| is more accurate.
|
| Llms getting better != a path to AGI.
| metalman wrote:
| the same is true of solar eclipses, there are no partial
| eclipses, untill the very last instant of the moon
| covering the sun, it is far to bright to look at, and
| then the stars come out and the solar flares are visible,
| the birds sing there evening songs and here I have told
| you of it, but at best it will be hint. AI is worse, much
| worse, as we have our own inteligence, but cant even
| offer a hint of where that line is and how to cross it,
| where to go to see it
| SalmoShalazar wrote:
| One could flip your post to say "AGI is coming" and be claiming
| the opposite, and it would be equally lacking insight. This is
| not "critical" to remember.
|
| There are interesting and well thought out arguments for why
| the AGI is not coming with the current state of technology,
| dismissing those arguments as propaganda/clickbait is not
| warranted. Yannic is also an AI professional and expert, not
| one to be offhandedly dismissed because you don't like the
| messaging.
| TheCraiggers wrote:
| I doing think that's fair to the person you replied to. At no
| time did they say they didn't like/dislike the message.
| Merely that there's a market for it, and thus, people may be
| biased.
|
| Telling us all to remember that there's potential for bias
| isn't so bad. It's a hot button issue.
| TheOtherHobbes wrote:
| You can remove the "not" and everything you wrote is just as
| true. If not more so.
|
| It's not the AGI sceptics who are getting $500bn valuations.
| d4rkn0d3z wrote:
| Oddly, in a bubble the highest valuations come just before
| the burst. This is obvious mathematical certainty that can be
| read by anyone viewing an exponential growth curve.
| kbrkbr wrote:
| > obvious mathematical certainty
|
| You mean like euclidean geometry?
| d4rkn0d3z wrote:
| I mean that as valuations rise before a bubble bursts,
| the curve provides that successive values grow, sometimes
| at an increasing rate. The greatest values, and change in
| values comes just before the bubble bursts. The point
| being that high valuations and increasing valuations are
| not very capable of distinguishing bubble/non-bubble. In
| fact, tulips were most valuable, and those values were
| climbing at the highest rate before the tulip bubble
| burst.
| kbrkbr wrote:
| Look, I asked
|
| >> obvious mathematical certainty
|
| > You mean like euclidean geometry
|
| To which you reply stuff about values and bubble bursts.
|
| What you said may or may not be true. But it's hard for
| me to tell how it is related to what I asked.
| d4rkn0d3z wrote:
| I said that high valuation or valuations rising at an
| increasing rate is common to bubbles. This clearly refers
| back to the comment about the current high and rising
| valuations for AGI prospects.
| kbrkbr wrote:
| I see that. I just have a hard time seeing how it is
| related to my question. Euclidean geometry was an
| "obvious mathematical certainty". Until it wasn't, and we
| learned that there are no "obvious mathematical
| certainties".
|
| So I wanted to understand what you mean by this concept.
|
| But we can leave it. It's not important. No harm meant.
| toasterlovin wrote:
| Right, it's exactly the opposite. What is the AI skeptic
| version of MIRI, for instance?
| d4rkn0d3z wrote:
| Can you say non-sequitor.
| drdeca wrote:
| I guess the AI skeptic version of MIRI would be like, an
| organization that tries to anticipate possible future large
| problems that could arise from people anticipating an AGI
| that never arrives, but which people might believe has
| arrived, and proposes methods to attempt to prevent or
| mitigate those potential problems?
| camillomiller wrote:
| By this measure, considering the current capex all over the
| board, there is a lot more incentive in pushing the "AGI IS
| NEAR AND WE AINT READY" narrative than the opposite. If AGI
| won't come, as it's highly probable, these companies are bust
| for billions and billions...
| good_stuffs wrote:
| Nobody even knows what AGI even is. This will most likely be
| defined by a corporation, not science. Due to obvious
| incentives.
| asimpletune wrote:
| AGI is being able to learn from first principles, not being
| trained on trillions of examples. If you can start with
| priors and perform a demonstration that is not already a
| prior then you are intelligent. It's the difference between
| the result and understanding the process that produces the
| result.
| kbrkbr wrote:
| Can you elaborate on the process by which you created this
| definition?
| asimpletune wrote:
| I just applied some stuff that's been known for a long
| time in a different context. But let me give you a
| scenario to think about as a demonstration of how I mean.
|
| Imagine we trained an AI on everything ever written, but
| the catch is we've restricted the training data to the
| year, let's say, 400 BCE and earlier (Ignore the fact
| that most of what was written then is lost to us now and
| just pretend that's not issue for our thought experiment)
| The AI is also programmed to seek new knowledge based off
| that starting knowledge.
|
| Also pretend that this AI has an oracle it could talk to
| that would help the AI simulate experiments. So the AI
| could ask questions and get answers but only in a way
| that builds ever so slightly off what it already knows.
|
| Making any progress at all in this experiment and
| discovering new knowledge is what we're after and "new
| knowledge" would be defined as some r that's demonstrated
| using some p and q as propositions, where r is neither p
| or q, and r is also correct in terms of 2025 knowledge.
|
| If the AI, with the aid of the knowledge it started with
| and the help of the oracle to let it ask questions about
| the world and build off that knowledge, can ever arrive,
| or exceed, 2025 knowledge then it's at least generally
| intelligent and equal to a human. Although the bar could
| maybe be even less.
|
| It loses, however, if it never advances, gets stuck in a
| loop, or in some other sense can't make progress.
|
| This is intelligence: to proceed from things everyone
| agrees on and ask questions and formulate assertions that
| depend on propositions holding true, and in the process
| demonstrate new things that were not already part of
| common belief.
|
| I don't know how this experiment could be done in real
| life with an LLM for example but this is a story version
| of what I mean in my original comment.
| kbrkbr wrote:
| > I just applied some stuff that's been known for a long
| time in a different context.
|
| I tried this, and screamed "Goal!!!!" on a golf course
| when the ball went in the hole. Did not work as expected.
| Lost my membership.
|
| But maybe I applied it wrong.
| politelemon wrote:
| What products are people building on not-AGI?
| gls2ro wrote:
| Usually the burden of proof is on the one who is making a
| positive claim like: AGI is here or even AGI is coming.
|
| The default position that does not need any more justification
| is the one that is skeptic or even agnostic to the claim that
| is made until proof is shown.
|
| So when talking about evidence as a way to prove a claim: AGI
| is coming is the team that needs to provide this evidence.
| Someone saying AGI is not coming can add as many arguments or
| opinions as they like but it does not usually invite to such a
| high scrutiny as saying they need to provide evidence.
| lostmsu wrote:
| Using the very basic definition of AGI where "general" is
| just cross-domain, as in e.g. chemistry and law, the very
| first ChatGPT was already it. Not very smart one though.
|
| "Modern" definitions that include non-intelligence related
| stuff like agency sound like goalpost moving, so it's unclear
| why would you want them.
| ivan_gammel wrote:
| Depends on definition of "cross-domain". Can any of the
| current models be plugged in in an emulator of a human with
| equivalent senses (vision, hearing etc) and process those
| inputs in real time with the same speed of reaction, i.e.
| emulate human or animal intelligence in interactions with
| environment? That would be truly cross-domain.
| lostmsu wrote:
| I defined cross-domain with an example. ChatGPT is not
| trained to practice chemistry and law, yet it can do
| both. It is cross-domain.
|
| You can make it stronger at being cross domain, but it
| satisfies the minimum requirement.
| ivan_gammel wrote:
| It cannot. It doesn't reason. Gambling and winning
| (1-10^N)*100% times is not the same as reasoning and
| providing accurate answers the same amount of times. If
| you reason about something, your errors fall into certain
| categories of fallacies often related to incomplete
| information. LLM hallucinations are easy to spot with
| reasoning, they are of statistical nature.
| lostmsu wrote:
| > (1-10^N)*100% times
| RugnirViking wrote:
| It is trained to practice chemistry and law. The reason
| it can do those because it's trained on an appreciable
| portion of all human output on both or those fields. If
| that's not training on them I don't know what is.
| lostmsu wrote:
| > It is trained to practice chemistry and law
|
| No, it is not. It is trained to predict next token, and
| it is trained to follow user instructions.
| nkrisc wrote:
| Faster than light travel is not coming.
|
| Given that AGI does not exist, "AGI is not coming" is the
| status quo until someone disproves it.
| vlan121 wrote:
| The goal of economic is not to reach AGI. It would solve the
| problems we have with the current market, therefore would it make
| less money, then to just "chase" for the AGI. Shirky principle in
| a nutshell.
| rationalpath wrote:
| Feels like we're all just betting on the biggest "what if" in
| history.
| rossdavidh wrote:
| On the one hand, that isn't necessarily a problem. It can be just
| a useful algorithm for tool calling or whatever.
|
| On the other hand, if you're telling your investors that AGI is
| about two years away, then you can only do that for a few years.
| Rumor has it that such claims were made? Hopefully no big
| investors actually believed that.
|
| The real question to be asking is, based on current applications
| of LLMs, can one pay for the hardware to sustain it? The
| comparison to smartphones is apt; by the time we got to the
| "Samsung Galaxy" phase, where only incremental improvements were
| coming, the industry was making a profit on each phone sold. Are
| any of the big LLMs actually profitable yet? And if they are, do
| they have any way to keep the DeepSeeks of the world from taking
| it away?
|
| What happens if you built your business on a service that turns
| out to be hugely expensive to run and not profitable?
| Salgat wrote:
| >On the other hand, if you're telling your investors that AGI
| is about two years away, then you can only do that for a few
| years.
|
| Musk has been doing this with autonomous driving since 2015.
| Machine learning has enough hype surrounding it that you have
| to embellish to keep up with every other company's ridiculous
| claims.
| ivan_gammel wrote:
| I doubt this was the main driver for the investors. People
| were buying Tesla even without it.
|
| Whether there is hype or not, the laws of money remain the
| same. If you invest and don't get expected returns, you will
| be eventually concerned and will do something about it.
| coolThingsFirst wrote:
| Why are companies allowed to lie? I really can't understand. If
| a person lies they lose credibility but it doesn't apply to the
| rich and powerful.
| bloppe wrote:
| Lying to investors is illegal, and investors have incentive
| and means to sue if they think they were defrauded. The
| problem is proving it. I'm sure a lot of founders genuinely
| believe AGI is about to appear out of thin air, so they're
| technically not lying. Even the cynical ones who say whatever
| they think investors want to hear are hard to catch in a lie.
| It's not really about being rich and powerful. That's just
| the unfortunate reality of rhetoric.
| orionsbelt wrote:
| Predictions about the future and puffery are not illegal.
| Lying about facts are. Nobody knows how far away AGI is,
| everyone just has their own predictions.
| xenotux wrote:
| It's not a crime to be wrong; it's only a crime to
| deliberately lie. And unless there's an email saying "haha
| we're lying to our investors", it's just not easy to prove.
| mmphosis wrote:
| mayhaps a prediction by an Artificial General Intelligence that
| is already here
| maxhille wrote:
| I mean there are different definitions on what to call an AGI.
| Most of the time people don't specify which one they use.
|
| For me an AGI would mean truly at least human level as in "this
| clearly has a consciousness paired with knowledge", a.k.a. a
| person. In that case, what do the investors expect? Some sort of
| slave market of virtual people to exploit?
| nemomarx wrote:
| Being able to make arbitrary duplicates of slaves would be
| profitable, as long as the energy and compute is lower than
| salaries yeah
| kbrkbr wrote:
| Investors don't use this definition. For one because it
| contains something you cannot measure yet: consciousness.
|
| How to find out if something has probably consciousness? Much
| less clearly? What is consciousness?
| seanalltogether wrote:
| Do we have a reasonable definition for what intelligence is? Is
| it like defining porn, you just know it when you see it?
| erikerikson wrote:
| One of my favorites is efficient cross domain maximization
| optimalsolver wrote:
| Efficient, cross-domain optimization.
|
| I believe that's Eliezer Yudkowsky's definition.
| erikerikson wrote:
| I did initially encounter it on LessWrong and modified it
| slightly according to my preference. Did he coin the term?
| There are a lot of ideas (not inappropriately) presented
| without attribution in that context.
| optimalsolver wrote:
| As far as I know, it's his own term.
| alan-crowe wrote:
| LLMs have demonstrated that "intelligence" is a broad umbrella
| term that covers a variety of very different things.
|
| Think about this story
| https://news.ycombinator.com/item?id=44845442
|
| Med-Gemini is clearly intelligent, but equally clearly it is an
| inhuman intelligence with different failure modes from human
| intelligence.
|
| If we say Med-Gemini is not intelligent, we will end up having
| to concede that actually it is intelligent. And the danger of
| this concession is that we will under-estimate how different it
| is from human intelligence and then get caught out by inhuman
| failures.
| AndrewDucker wrote:
| My personal definition is "The ability to form models from
| observations and extrapolate from them."
|
| LLMs are great at forming models of language from observations
| of language and extrapolating language constructs from them.
| But to get _general_ intelligence we 're going to have to let
| an AI build their models from direct measurements of reality.
| daveguy wrote:
| > LLMs are great at forming models of language
|
| They really aren't even great at _forming models_ of
| language. They are a single model of language. They don 't
| build models, much less use those models. See, for example,
| ARC-AGI 1 and 2. They only performed ARC 1 decently [0] with
| additional training, and are failing miserably on ARC 2.
| That's not even getting to ARC 3.
|
| [0] https://arcprize.org/blog/oai-o3-pub-breakthrough
|
| > Note on "tuned": OpenAI shared they trained the o3 we
| tested on 75% of the Public Training set. They have not
| shared more details. We have not yet tested the ARC-untrained
| model to understand how much of the performance is due to
| ARC-AGI data.
|
| ... Clearly not able to reason about the problems without
| additional training. And no indication that the additional
| training didn't include some feature extraction, scaffolding,
| RLHF, etc created by human intelligence. Impressive that fine
| tuning can get >85%, but it's still additional human directed
| training and not self contained intelligence at the level of
| performance reported. The blog was very generous making the
| undefined "fine tuning" a footnote and praising the results
| as if they were directly from the model that would have cost
| > $65,000 to run.
|
| Edit: to be clear, I understand LLMs are a huge leap forward
| in AI research and possibly the first models that can provide
| useful results _across multiple domains_ without being
| retrained. But they 're still not creating their own models,
| even of language.
| pan69 wrote:
| > Is it like defining porn
|
| I guess when it comes to the definition of intelligence, just
| like porn, different people have different levels of tolerance.
| d4rkn0d3z wrote:
| To share my experience, 25 years ago I looked into AI and had
| inclinations to see what scaling compute would do. It took no
| time to find advisors who told me the whole program I had in mind
| could not gain ethics approval and was mathematically limited.
| The former road block is now lifted due to the fact that nobody
| cares about ethics any more, the latter seems to be the remaining
| hurdle.
| lif wrote:
| so, without sarcasm: how many data centers is this non-happening
| worth? in other words, what justifies the huge spend?
| dang wrote:
| This is more interesting and deserves better discussion than we
| got from the previous title, which was derailed by the "AGI" bit,
| so I replaced the title with a representative sentence from the
| video.
|
| (Edit: plus a question mark, as we sometimes do with contentious
| titles.)
| tim333 wrote:
| >The era of boundary-breaking advancements is over
|
| Maybe for LLMs but they are not the only possible algorithm. Only
| this week we had Genie 3 as in:
|
| >The Surprising Leap in AI: How Genie 3's World Model Redefines
| Synthetic Reality https://www.msn.com/en-us/news/technology/the-
| surprising-lea...
|
| and:
|
| >DeepMind thinks its new Genie 3 world model presents a stepping
| stone toward AGI https://techcrunch.com/2025/08/05/deepmind-
| thinks-genie-3-wo...
| dgs_sgd wrote:
| How different are world models from LLMs? I'm not in the AI
| space but follow it here. I always assumed they belonged to the
| same "family" of tech and were more similar than different.
|
| But are they sufficiently different that stalling progress in
| one doesn't imply stalling progress in the other?
| tim333 wrote:
| There are similarities with that one. From their website:
|
| >It is comprised of a spatiotemporal video tokenizer, an
| autoregressive dynamics model, and a simple and scalable
| latent action model.
|
| my point is more people can try different models and
| algorithms rather than having to stick to LLMs.
___________________________________________________________________
(page generated 2025-08-09 23:01 UTC)