captainclam

I can imagine that one of the aspects of being social creatures is that we are drawn to the entropy/surprise factor of natural language produced by a brain that isn't our own. Deriving a shared understanding without being able to precisely model the language output of someone else is like, the fundamental substance of conversation.

Maybe that's the key ingredient that's missing from LLM-generated text. Even when there is actual content there, there's this lack of surprise / uncertainty in its delivery that we require to stay engaged. We may not know what will be communicated, but knowing exactly how it will be communicated just kills our innate interest.

Obviously the kid would choose the colorful modern game over some ancient NES game. They would literally choose one of those weird phone games you see crazy ads for over Super Mario Bros.

> "minecraft in skyrim" is cool for about 10 seconds till you think about actually sitting down to play it.

Yeah, it is kinda wild how the infinitude of possibility suddenly makes everything so dull.

> And the thing is, it's not the technology itself, which could be used to produce meaningful art.

Exactly. I am not de-facto against the use of AI in art. But artful use of the medium is sparse at the moment.

> movies are going to become video games and video games are going to become something unimaginably better

God what an annoying thing to say. Suggesting that movies will essentially become games, or that future video games will be unimaginably better (presumably because they are driven by generative world models) betrays such a surface level appreciation for what makes art and culture and all this humanity so beautiful as it is.

The entire point of movies is the deliberate choices of the filmmakers! Movies will not be improved by our ability to choose what the main character looks like or add a style filter. Super Mario 64 is great because it is an expression of the very hardware that constrains it. Inevitable future versions where anything can happen because someone hooked the ROM up to a generative world model won't be better for it.

I agree that this technology is absolutely world changing, and am cautiously eager to see what comes of it, but it just bums me out, perhaps needlessly, that the people at the helm can be so artless, unreflective. Whatever.

100%.

And a corollary being that without any time spent on the "mundane", almost every single aspect of the work becomes the genuinely consequential decision-making steps.

Which I find exhausting.

I completely understand where you're coming from. While understanding the math and architectures of these large models can be edifying, it lacks some of the liberating/empowering spirit of traditional software. You can build a bespoke to-do app that is better suited to your needs than anything produced by some productivity software company, but you're totally beholden to one of the few big players for access to an effective LLM.

But there is so, so much interesting work and research to do at a much smaller scale. I work on drug discovery models, and none of my work requires data or infrastructure that are not generally accessible to the public (it's nice to have the computing resources of my institution but there are other ways to access a GPU).

So if you are really interested in this stuff, I would suggest looking at the endless supply of interesting problems in domains that are not dominated by huge frontier labs.

"The agent could write refund code all day long. What it can’t do is decide what your company owes a customer whose card has expired..."

In 2031: Why on earth not?

These are good points, though OP did specifically say that Meta is going to win "consumer" AI. I don't think the average consumer of AI is going to ever have to think about harnesses or worry about SOTA, they really will continue to interact with whatever is stickiest.

The following certainly does not describe all mathematics or all LLM operations, but its my mental model for why LLMs are so effective in several fields.

Lots of mathematics entails recognizing that a tool/techniques that works in one area can be adapted to a completely different problem. For a human to do this requires having a robust mental model of the technique and the problem itself, and probably having just the "right" mental model to see how they fit together. And we are talking about extremely hard/sophisticated techniques and extremely hard/sophisticated problems, so the ability to deeply appreciate either can require years of study. And on top of that, even if the mathematician has sufficient understanding to develop the intuition that the technique might work, they will need profound fortitude, skill, and patience to see the solution through.

Unlike human mathematicians, who will develop a deep understanding of only a relative handful of techniques and problems in their lifetime, LLMs have ALL written techniques, problems, solutions, etc stored in their internal representations. So when they ably match a problem to a technique, we can imagine the operation as a supercharged autocomplete, or even an extremely proficient search engine. Their stroke of genius is really a consequence of capacity, moreso than cleverness.

Huh. I guess I don't agree that this would at all characterize Tao's relationship with AI. I haven't read a single thing from him to suggest that he is "seething" over AI's remarkable capabilities. In fact he's a prolific user of AI in mathematics research. I don't think the scene serves a useful analogy.

Also Tao's reference to "feeling for" Salieri makes a lot more sense, the audience is clearly supposed to feel for him! That's what makes the scene effective.

And why would I want to see AI humiliate Tao?

I'm sorry if I misunderstood your comment. Thank you for the link, it is useful context.

I replied to a different comment with an explanation as to why I do not perceive Tao as a gatekeeper; in short that his current beat seems to be entirely about expanding understanding and access to math by improving the communication of opaque proofs, which we will shortly have a near infinite supply of due to AI.

I will also mention that he maintains a realistic, clear eyed attitude with respect to the inevitability of AI in math, while being sensitive to its disruptive nature in his field. Everything here strikes me as constructing a thoughtful, grounded way forward for his community.

I'm sure there will be missteps along the way, but I think Tao is overall an excellent example of maintaining a balanced position between the poles of absolutism that seem to capture so many.

I am curious where you interpret this as gatekeeping behavior.

I was not familiar with the mathathon, and I suppose I am ignorant of the circumstances in which Tao allegedly bullied students into changing the event. I don't personally have any reason to doubt the veracity of the "Joint Statement about Mathathon", but perhaps you have some insight that I don't. I will say that I would imagine that any input from Tao, even in passing, probably has outsized influence on the organizational direction of the event due to his preeminence, but I'm not so sure that necessarily rises to the level of bullying.

But from what I gather, the theme of the statement is to expand the scope of participation to include work that is quite literally the /opposite/ of gatekeeping, i.e. taking proofs that are popularly understood to be opaque or difficult to understand (from any source, AI, human, or otherwise) and developing a new way to communicate them that is more accessible to more people.

I do see how it may come off as a slight to the original mathematicians who developed those proofs. But developing new ways to communicate already established concepts should not diminish the original works. There are probably much better communicators/teachers of relativity than Einstein ever was. No one out there thinks any less of Einstein because of Sean Carrol's "Spacetime and Geometry" textbook.

And this to me reads as only embracing the use of AI in mathematics, which is not gatekeeping behavior. Tao continues to emphasize that once proofs become effectively automatable, deeply comprehending and being able to transmit that understanding to others will become an increasingly important aspect of the field, the natural way for mathematicians to use the time that was once occupied by developing the proofs themselves.

Literally all of this is about expanding the accessibility of mathematics, which once again, I perceive to be quite the opposite of gatekeeping. Maybe we have different definitions of that term.

Also I have not seen Amadeus, so no comment about Salieri.

I also work in drug discovery, and I completely agree; the extent to which the actual causal mechanisms of the efficacy of many drugs is woefully short of some human-appreciable first-principles based explanation. The point is especially undercut by hypothetically suggesting that this drug /does/ in fact pass stage 3 trials, at which point I can't fathom anyone would have a problem advancing it.

Really wish he had chosen a different example; this particular bullet point has been making the rounds on X/twitter to paint Tao as an example of some sort of gatekeeping luddite who would deny the world a post-abundance future in order to maintain the prestige of his particular career path...which is tough because I cannot think of a more responsible steward of our inevitable AI future than Tao at the moment.

"If every product is used by an agent driving a CLI, what’s going to differentiate Basecamp or Fizzy from the cheapest alternative?"

At that point, why even bother with "driving a CLI"? I have a hard time seeing how this all doesn't go away soon. At the current trajectory, I am not seeing a future where software like Basecamp or Fizzy or the cheapest alternative are competitive with "Claude, build a basecamp-style project management tool for my team."

I don't mean to say that thoughtfully built, opinionated software doesn't have intrinsic value, I believe it will always be "better" in certain aspects...I just can't fathom that a market for it will exist in very short order.

I would LOVE to be talked out of this perspective.

I imagine a good model of the brain would contribute enormously to alleviating psychiatric/mental health disorders.

What's the tell? It doesn't seem AI generated to me but I may not be a great judge.

I don't think the quality of discussion about social media suffers from lack of specification. Whether or not you consider HN to be social media, or wherever your decision boundary is, doesn't change that most of the conversation does apply to the general class of apps/websites that have become de-facto short form video platforms. Which lots of people use, so the effects of use are consequential and worth discussing.

The conversations on consciousness though...oh man. I have to steel myself before diving into that mess.

I use the firefox extension "Unhook" to completely hide suggested content on Youtube. Really effective, I kindof can't believe how much time I spent getting suckered into watching video essays that absolutely did not deliver.

Just to be clear, your comment had a general statement about how you perceived the motivations of "people buying vinyl". That's what I was responding to. (People using VHS filters on social media is by definition social signalling so no comment there lol).

And I completely agree with your point about touting film as "easier" than digital. That's a stretch.

There are plenty of people who sincerely enjoy the aspects that make older tech less convenient or practical. Maybe it's an appreciation for the engineering or "comprehensibility," often it's because older tech produces unique outputs that can't be adequately captured by newer technology.

Reducing people's interest to "social signalling" comes off as dismissive.

Not asking adversarially at all here: what do you mean by resisting with "real numbers" without media campaigns, social media, or protesting? What do the vested parties actually do to secure their second amendment rights? Do you just mean having large voting blocs?

Author really should have figured out a better word than "vanity."

It looks to me like OpenAI's image pipeline takes an image as input, derives the semantic details, and then essentially regenerates an entirely new image based on the "description" obtained from the input image.

Even Sam Altman's "Ghiblified" twitter avatar looks nothing like him (at least to me).

Other models seem much more able to operate directly on the input image.

You must not end up reading much scientific literature then.

lol

The two dogs I know that share this behavior are border collies.

The seahorse emoji is one of the canonical "Mandela effects". These are things that a large group of people collectively (mis)remember, but turn out to have never existed. Classic examples include the cornucopia in the Fruit of the Loom label (never there), and the wording on car mirrors "objects in the mirror may be closer than they appear." (There's no record of 'may be closer', just 'are closer').

Unfortunately, the discussion around Mandela effects gets tainted by lots of people being so sure of their memory that the only explanation must be fantastical (the timeline has shifted!), giving the topic a valence of crazy that discourages engagement. I find these mass mis-rememberings fascinating from a psychological perspective, and lacking satisfying explanation (there probably isn't one).

So here we're seeing LLMs "experiencing" the same mandela effect that afflicts so many people, and I sincerely wonder why? The obvious answer is that the training data has lots of discussions about this particular mandela effect, ie people posting online "where is the seahorse emoji"? But those discussions are probably necessarily coupled with language that ascertains 'no, the seahorse emoji does not exist.' That's why the discussion is there in the first place! so why does the model take on the persona of someone that is sure it does exist? Why does it steer the models into such a weird feedback loop?

I've always been surprised by the official homeless population count, but it turns out there's a lot more to it.

The department of HUD generates this ~771K figure from a "point-in-time" estimate, a single count from a single night performed in January. They literally have volunteers go out, count the number of homeless people they observe, and report their findings.

It's not hard to imagine why this is probably a significant undercount. There is likely a long tail of people that happened to be in a situation that night where they were not able to be counted (i.e. somewhere secluded, sleeping in a friend's private residence that night, etc).

Even if these numbers are correct, to my mind a "crisis" is still more characterized by the trend than the numbers in absolute. From the first link you provided, we saw a 39% increase in "people in families" experiencing homelessness, and 9% in individuals. A resource from the HUD itself suggests a 33% increase in homelessness from 2020-2024, 18% increase from 2023-2024. That is far apace of the population increase in general.

https://www.huduser.gov/portal/sites/default/files/pdf/2024-...

And even then, I would say many people would suggest that the change in visible homelessness they've experienced in the last 10 years would amount to "crisis" levels, at least relative to the past.

It's completely fair to argue that it is not in fact a crisis, but claiming that it is certainly not "baseless."

Wow, there really is an xkcd for everything.