captainclam

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.

My read is that the author is saying it would have been really nice if there had been a really good protocol for storing data in a rich semantically structured way and everyone had been really really good at adhering to that standard.

Is that the main thrust of it?

It's very easy to imagine a world where all these things are solved, but it is a worse world to live in overall.

I don't think it is "bad" to be sincerely worried that the current trajectory of AI progress represents this trade.

To be clear, I'm pretty sure the half-trillion figure is the projected combined investment between SoftBank, OpenAI, Oracle, and MGX. Not public, US tax-payer dollars.

If that's not what you meant my apologies. Reason I'm quick to point this out is I think some of the writing/headlines around are suggestive of this misconception.

Sounds like there's not enough tax revenue!

This is one of the many many experiences in the tapestry of people figuring out how to use this new tool.

There will be many such cases of engineers losing their edge.

There will be many cases of engineers skillfully wielding LLMs and growing as a result.

There will be many cases of hobbyists becoming empowered to build new things.

There will be many cases of SWEs getting lazy and building up huge, messy, intractable code bases.

I enjoy reading from all these perspectives. I am tired of sweeping statements like "AI is Making Developers Dumb."

Exactly. If the whole "deep research" thing pans out, and we have models that can reliably produce proper literature reviews in 10 minutes...that alone will be an enormous boon to research.

Then add all the practical/mundane tasks that you mentioned, and you've got quite the multiplier.

Crucially, this doesn't just require noise but it requires "taste."

I tend to fall back on music creation as an example of this notion. Lots of innovation in music is experimentation/exploration of "noise," (not necessarily literal white noise) but requires the ear of a discerning musician who ultimately goes "Ooh! I liked that" or passes a "generated sample" by.

This is where I wonder if LLMs can ever innovate. I'm not sure they can develop "taste" for things outside of their distribution. However, I could just as easily be convinced that humans can't either, and sophisticated "taste" is just the exploration of obscure regions of the combinatorial space generated from previously observed samples!

Especially with the proliferation of generative AI, I anticipate something of a tech backlash in the next decade, and performatively NOT looking at one's phone will be part of it.

I'm sure that this already exists to some extant in certain subgroups, but I'd bet a small amount of money that this will grow to be a visible trend.

Just a fun thought!

Definitely interesting, but I'm not so sure that such a study can yet make strong claims about AI-based work in general.

These are scientists that have cultivated a particular workflow/work habits over years, even decades. To a significant extent, I'm sure their workflow is shaped by what they find fulfilling.

That they report less fulfillment when tasked with working under a new methodology, especially one that they feel little to no mastery over, is not terribly surprising.

What is dogfooding?

"Elon Musk Ally Tells Staff ‘AI-First’ Is the Future of Key Government Agency" from Wired

https://archive.is/jyFCy

This isn't unequivocal proof, but the broad goal automation lends itself pretty strongly to LLMs, and oh boy what LLM technology do you think they want to use.

semantics shmemantics.

> This was one of the rather many areas where Star Trek failed to really consider the implications of its concepts, probably because it would simply break the world building.

This is true of pretty much all scifi! It's funny seeing super-futuristic depictions of star-fighter pilots and combatants with firearms and its just...so crushingly evident that humans will not have supremacy in these arenas very shortly.

Frank Herbert must have anticipated this complication and side-stepped the whole issue by preemptively canonizing the Butlerian Jihad.

Haha, I've had the same thoughts, that of course computers/AI/droids of that conversational capacity were conscious. You'd be a brute not to think that!

And all of a sudden, LLMs absolutely have the command of natural language that once seemed such an obvious indicator of sentience, and now I find myself one of those bigots who don't believe in robot rights!

I'm being silly, but I do think there are implications here with respect to the future debate on AI sentience. I guess I once thought there would be this threshold where the reality of an AI's inner experience became blatantly obvious, but I see now that this is going to be a profoundly thorny problem.

Who knows, maybe in several decades we'll have a consciousness-o-meter that demonstrates that LLMs have had some degree of awareness all along.

Oh never mind then.

I am sincerely curious about your thoughts on the origins of mitochondria, and how you came to feel so strongly about it.

Pens are all well and good, but what are people's favorite notebooks? I've found that the paper is perhaps just as important as the pen in a satisfactory writing experience.

Hmmm...this wasn't at all my point, but I understand why I my comment could be read this way (the failure to communicate is mine).

I am not suggesting an equivalency between all efforts at all levels, or that innate talent doesn't exist. I was not at all speculating about the nature of preternatural talent. The expertise-gap I was invoking was more like early undergrad to late grad school levels, not pick-up football to Premier League.

The concept I'm getting at is that of a mental block I have observed in myself, and I suspect resides in others, which is really subtle but ultimately quite limiting. I'd have to think harder than I want to at the moment if I were to try to articulate it more clearly, but I do want to be clear that my comment wasn't about innate talent.

A really fascinating corollary I've observed since I've gotten into more advanced maths, or even doing actual research as a PhD student, is that there's nothing special going on at the higher levels. You're just working with different materials. Materials that require more time and effort to 'get', but once you get them they are just another tool at your disposal.

I was similar to the author in that, throughout high school and undergrad, I presumed that the mind that could comprehend advanced math or do novel research (in any field) was truly unknowable. Like there was this x-factor they had that wasn't there for me.

I've long enjoyed puzzle games (like The Witness or Stephen's Sausage Roll). It turns out that problem solving in non-trivial domains is never terribly different than problem solving in those games, or any other domain really. Like my brain isn't doing anything different than the usual tree-search algorithm that any chess player performs when they are projecting moves ahead into the future.

Its just iterating on concepts that seem abstruse to most people. But at the end of the day, deep problem solving in math or AI research tends to be the same moving-shapes-around-in-my head that I would do if I was trying to move an awkwardly shaped couch through a narrow doorway.

"Classic Wolfram — brilliant, reimplements / comes at a current topic using only cellular automata, and draws some fairly deep philosophical conclusions that are pretty intriguing."

Wolfram has a hammer and sees everything as a nail. But its a really interesting hammer.

“Mimicry of mechanisms left by evolution” would be a much less parsimonious explanation than just having those mechanisms, i.e. Occam’s razor.

Addressing your first thought…anything that you would call “objective” can be “doubted” by ceding the tiny tiny possibility that you are a simulation or Boltzmann brain or brain in a vat. The evidence before you may not actually be representative of the “objective” reality.

The fact that there is experience at all, the contents of which may be “doubted”, cannot be doubted.

I’m not unequivocally claiming this but that’s the thrust of the argument.

Op writes "I’m writing this comment so that people who want to know more about alternative theories of consciousness (to materialism/physicalism [1]) can know where to go to find well-argued positions on the topic."

They very specifically state that these people are good points of entry for "well-argued positions on the topic." Linking to specific literature would have been better, but this isn't "materialism/physicalism is wrong because of these people's credentials."

I guess you could say I don't know RWP from Adam! :D

My og comment wasn't to accurately explain gradient optimization, I was just expressing a sentiment not especially aimed at experts and not especially requiring details.

Though I'm afraid I subjected you to the same "cringe" I experience when I read pop sci/tech articles describe deep learning optimization as "the algorithm" being "rewarded" or "punished," haha.

Ugh, exactly, it's so cool. I've been a deep learning practitioner for ~3 years now, and I feel like this notion has really been impressed upon me only recently.

I've spent an awful lot of mental energy trying to conceive of how these things work, when really it comes down to "does increasing this parameter improve the performance on this task? Yes? Move the dial up a bit. No? Down a bit..." x 1e9.

And the cool part is that this yields such rich, interesting, sometimes even useful, structures!

I like to think of this cognitive primitive as the analogue to the idea that thermodynamics is just the sum of particles bumping into each other. At the end of the day, that really is just it, but the collective behavior is something else entirely.