aithrowawaycomm

> You can say LLMs are fundamentally dumb because of their inherent linearity. Are they? Isn’t language by itself linear (more precisely, the presentation of it)?

Any linearity (or at least partial ordering) of intelligence comes from time and causality, not language - in fact the linearity of language is a limitation human cognition struggles to fight against.

I think this is where "chimpanzees are intelligent" comes to the rescue - AI has a nasty habit of focusing too much on humans. It is vacuous to think that chimpanzee intelligence can be reduced to a linear sequence of oohs-and-aahs, although I suspect a transformer trained on thousands of hours of chimp vocalizations could keep a real chimp busy for a long time. Ape cognition is much deeper and more mysterious: imperfect "axioms" and "algorithms" about space, time, numbers, object-ness, identifying other intelligences, etc, seem to be somehow built-in, and all apes seem to share deep cognitive tools like self-reflection, estimating the cognitive complexity of a task, robust quantitative reasoning, and so on. Nor does it really make sense to hand-wave about "evolutionary training data" - there are stark micro- and macro-architectural differences between primate brains and squirrel brains. Not to mention that all species have the exact same amount of data - if it was just about millions of years, why are bees and octopi uniquely intelligent among invertebrates? Why aren't there any chimpanzee-level squirrels? Rather than twisting into knots about "high quality evolutionary data," it makes a lot more sense to point towards evolution pressuring the development of different brain architectures with stronger cognitive abilities. (Especially considering how rapidly modern human intelligence seems to evolved - much more easily explained by sudden favorable mutations vs stumbling into an East African data treasure trove.)

Human intelligence uses these "algorithms" + the more modern tool of language to reason about the world. I believe any AI system which starts with language and sensory input[1], then hopes to get causality/etc via Big Data is doomed to failure: it might be an exceptionally useful text generator/processor but there will be infinite families of text-based problems that toddlers can solve but the AI cannot.

[1] I also think sight-without-touch is doomed to failure, especially with video generation, but that's a different discussion. And AIs can somewhat cheat "touch" if they train extensively on a good video game engine (I see RDR2 is used a lot).

Newborns (and certainly toddlers) seem to understand the underlying concepts for these things when it comes to visual/hepatic object identification and "folk physics":

  A short list of abilities that cannot be performed by CompressARC includes:

  Assigning two colors to each other (see puzzle 0d3d703e)
  Repeating an operation in series many times (see puzzle 0a938d79)
  Counting/numbers (see puzzle ce9e57f2)
  Translation, rotation, reflections, rescaling, image duplication (see puzzles 0e206a2e, 5ad4f10b, and 2bcee788)
  Detecting topological properties such as connectivity (see puzzle 7b6016b9)

Note: I am not saying newborns can solve the corresponding ARC problems! The point is there is a lot of evidence that many of the concepts ARC-AGI is (allegedly) measuring are innate in humans, and maybe most animals; e.g. cockroaches can quickly identify connected/disconnected components when it comes to pathfinding. Again, not saying cockroaches can solve ARC :) OTOH even if orcas were smarter than humans they would struggle with ARC - it would be way too baffling and obtuse if your culture doesn't have the concept of written standardized tests. (I was solving state-mandated ARCish problems since elementary school.) This also applies to hunter-gatherers, and note the converse: if you plopped me down among the Khoisan in the Kalahari, they would think I was an ignorant moron. But it makes as much sense scientifically to say "human-level intelligence" entails "human-level hunter-gathering" instead of "human-level IQ problems."

I believe Feynman was basically mistaken about the second point, though of course the field was brand new at the time - it is certainly possible to simulate the measurements of quantum mechanics to arbitrarily high accuracy on a classical computer with pseudorandom number generation; if you replace the pseudorandomness with a physical random number generator, then it might even be formally equivalent to a quantum computer (I think that's an open question, haven't tracked the developments in a while).

"Negative probabilities" is not quite right - towards the end of his life Feynman wondered about generalizing probability but that was just about intermediate calculations: he declared physical events cannot have nonnegative probabilities (in the same sense that physically I can't have negative three apples, -3 is a nonphysical abstraction used to simplify accounting). Negative probabilities are not part of modern quantum mechanics, where probabilities are always nonnegative and sum to 1. Quantum states can have negative/complex amplitudes but the probabilities are positive (and classical computers are just as good/bad at complex arithmetic as they are any other).

The "hidden variables" comment makes me think Feynman was actually a bit confused about the philosophy of computation - a classical computer cannot simulate how a quantum particle "truly" evolves over time, but that's also the case for a classical particle! Ultimately it's just a bunch of assembly pushing electrons around, that has nothing to do with a ball rolling down a hill. Computers only have Schrodinger's equation or Newton's laws, which don't care how the motion "truly" works, they just care that the measurement at the end is correct. If a computer gets the correct measurements then by definition we say it simulates the phenomenon.

Edit: clarifying this last point, Newton’s laws do have a known “hidden variables” theory in the sense that we know how an ensemble of high-temperature quantum particles can “average out” into Newton’s laws, there is an electrostatic theory of mechanical contact, etc. This does not (and seemingly cannot) exist for quantum mechanics, but merely having a quantum computer wouldn’t by itself help us figure out what’s going on: the output of a quantum computer are the “visible” variables, aka the observables. The fact that quantum computers are truly using the non-observables, whatever those might be, seemingly cannot be experimentally distinguished from a sufficiently accurate classical computer doing numerical quantum mechanics. If it turns out there is experimentally a serious difference between the results of quantum computers and classical qubit simulators, that would suggest an inadequacy in the foundations of QM.

It is vacuously true that a Turing machine can simulate a human mind - this is the quantum Church-Turing thesis. Since a Turing machine can solve any arbitrary system of Schrodinger equations, it can solve the system describing every atom in the human body.[1]

The problem is that this might take more energy than the Sun for any physical computer. What is far less obvious is whether there exist any computable higher-order abstractions of the human mind that can be more feasibly implemented. Lots of layers to this - is there an easily computable model of neurons that encapsulates cognition, or do we have to model every protein and mRNA?

It may be analogous to integration: we can numerically integrate almost anything, but most functions are not symbolically integrable and most differential equations lack closed-form solutions. Maybe the only way to model human intelligence is "numerical."

In fact I suspect higher-order cognition is not Turing computable, though obviously I have no way of proving it. My issue is very general: Turing machines are symbolic, and one cannot define what a symbol actually is without using symbols - which means it cannot be defined at all. "Symbol" seems to be a primitive concept in humans, and I don't see how to transfer it to a Turing machine / ChatGPT reliably. Or, as a more minor point, our internal "common sense physics simulator" is qualitatively very powerful despite being quantitatively weak (the exact opposite of Sora/Veo/etc), which again does not seem amenable to a purely symbolic formulation: consider "if you blow the flame lightly it will flicker, if you blow hard it will go out." These symbols communicate the result without any insight into the computation.

[1] This doesn't have anything to do with Penrose's quantum consciousness stuff, it just assumes humans don't have metaphysical souls.

> These discontinuities are not just an artifact of poor implementation; it can be proven that any representation of 3D rotations using only three values must contain discontinuities.

This is a bit pedantic - and the blog post actually does clarify this - but the problem isn't that a 3D representation of representations has "discontinuities" as such, it's that it's not orientable in Euclidean 3D space. It is similar to the Klein bottle - the mathematical description is continuous, but any Klein bottle made of real-world glass has to intersect itself. Or likewise that a Mobius strip can be demonstrated in 3D but can't be built in Flatland without a 3D entity doing the copy-pasting. Reality just has the wrong topology to represent the full group of rotations. Hence the discussion about projective space later in the blog post.

So adding a fourth gimbal is really tantamount to a correctly-oriented embedding of 3D rotations onto a 4-torus (that is, [0,2pi]^4).

Gimbal lock also relates to another issue of continuity, related to the "plate dance."[1] Rotations themselves have a sense of continuity (infinitesimal differences in either the angle or axis of rotation, aka they are Lie groups), but Euler angles fail to respect the equivalent fundamental theorem of calculus: adding a bunch of infinitesimal changes might say you are at the identity rotation according to Euler angles, but in reality you have flipped the meaning of the right-hand rule and the overall state of the system is not at the identity. In a robotics context, the robot's hand might have done a complete rotation, but its arm is twisted without the robot "knowing." I believe robotic arms used to have a serious problem with this, either overrotating and breaking the arm, or swinging dangerously fast in the opposite direction. Using quaternions / a fourth gimbal / etc. there would be a measurable phase or pole indicating the true state of the system, and letting the robot know how to rotate its arm without malfunctioning. So, like the Apollo mission, the need for a fourth dimension to keep track of that stuff - and even the quaternion multiplication structure - comes about pretty naturally without ever thinking about abstract math.

[1] https://en.m.wikipedia.org/wiki/Plate_trick

That is the video I am talking about.

The only video I saw of this happening seemed to be in a text conversation where they were previously talking about politics. I don't know how Apple's transcription works but it might have jumped the gun simply because the guy talks about politics a lot + something about his voice.

I see no indication that this is an ubiquitous issue.

That is true but it seems like most authoritarians invent a vague emergency to justify grabbing specific powers (e.g. Trump declaring a "border emergency"), whereas a specific emergency for a specific event has a time limit.

In particular: getting Trump to rescind his border "emergency" will be a long slow process of organizing and accumulating public pressure, whereas there will be an overwhelming amount of public pressure to end the curfew as soon as the power is restored.

[flagged]

He is a bit older. Linear algebra is also very old, but it didn't really become the field we know today until the 1950s. I would add that in 2025 it is cheap to buy a computer that can solve large linear systems, but that certainly wasn't true in 1975, so linear algebra was less applicable in the real world.

I am not too familiar with the pedagogical history of linear algebra, but I've been reading some advanced undergraduate geometry texts from the 30s-60s and linear algebra was generally not an assumed prerequisite. There was a particular separation between the studies of "two and three dimensional vector spaces over R" (largely geometric) versus "finite dimensional vector spaces over a field" (entirely algebraic), and determinants were presented directly as volume computations. These days undergraduates mostly treat R^2 and R^3 algebraically, maybe at the expense of geometric understanding. (E.g. Euler's rotation theorem is easily proved when restated as a theorem about matrices over R^3 with determinant +1, but Euler's original statement and proof using spherical trigonometry is deeper.)

No, it is not your right to copy! If you want to keep using euros and have free travel to other Eurozone countries, you need to respect EU law. The tax does not respect EU law because it makes no effort to actually offset losses to copyright holders.

Your comment doesn't even make sense! The intention was never to make piracy okay, it was to offset a tiny fraction of the financial losses. These financial losses are still occurring even now that piracy is formally illegal, so the tax is still justified.

More generally this seems a form of freeloading: picking the parts of the EU that you like, ignoring the parts that don't. Who cares about society, it's just me me me. Pure selfishness.

It certainly has!

https://stichtingbrein.nl/largest-ever-criminal-action-again...

https://piracymonitor.org/netherlands-odido-is-ordered-to-bl...

https://piracymonitor.org/netherlands-brein-reports-two-pira...

11 years ago the EU made the Netherlands change their position: https://www.zdnet.com/article/downloading-pirate-material-fi... AFAIK this is still the case - the Netherlands is more poorly-enforced than other EU countries, but it's still illegal to download pirated material.

I actually wonder if many individual cops use "null" (the ordinary word) as a shorthand for "plate not applicable," "not identified", "missing" etc - semantically the same thing as NULL in programming, but then in this case it wouldn't be a database NULL error. In theory the same thing could happen if someone had a DEALER vanity plate. (Though that choice might be rejected for obliquely referencing drugs.)

I think in 2025 we can do a bit better than "I scared the LLM into compliance": https://xcancel.com/colin_fraser/status/1892683791514194378

This is hilarious:

  Even those without the last name Null are finding themselves caught in the void. Joseph Tartaro got a license plate with the word “NULL” on it nearly 10 years ago. The 36-year-old security auditor thought it would be funny to drive around with the symbol for an empty value. Maybe a police officer who tried to give him a ticket would end up writing null into the system and not be able to process it, he joked to himself.

  In 2018 he paid a $35 parking ticket. Soon afterward, he said, his mailbox was flooded with hundreds of traffic tickets for incidents he hadn’t been involved in. Tickets were from other counties and cities for vehicles of different colors, makes and models. A database had associated the word “null” with his personal information and citations were sent to Tartaro, who lives in Los Angeles.

I think he means that you don't need to train a human to understand that a calculator is useful, and in particular when a problem is hard enough that you need to bust out a calculator. That sort of logic is self-apparent in humans, but struggles to be consistently evoked in LLMs.

That said, I was using simple +*-/ calculators as a small child and I don't think I needed to be taught anything other than MC/MR. The tool is intuitive if you are familiar with formal written arithmetic (of course hunter-gatherers couldn't make sense of it).

The most serious problem is that Elon Musk is the 2nd most powerful person in the federal government, and seems to be giving cabinet officials direct orders, yet he has not divested himself of his SpaceX shares.

I don't understand how anyone can reasonably pretend this is okay. Elon Musk + the birthright citizenship order + Eric Adams means Donald Trump is by far the most corrupt president in American history.

I expect SpaceX to use the same care, patience, humility, and honesty as DOGE:

Trump administration fires and then tries to rehire nuclear weapons workers in DOGE reversal https://www.cbsnews.com/news/doge-firings-us-nuclear-weapons...

USDA accidentally fired officials working on bird flu and is now trying to rehire them https://www.nbcnews.com/politics/doge/usda-accidentally-fire...

And some strong evidence that DOGE is rewriting government databases to cover their tracks: https://xcancel.com/electricfutures/status/18920196528358073... (Note how petty this is - DOGE appears to be falsifying data rather than publicly admitting they were wrong! I assume Elon is personally responsible, just like Trump personally drew on the map with a Sharpie.)

The most frustrating iteration of this is when someone uses an article to make a point in an argument...and they didn't read it! A clickbait headline seems to support their claim, but the substance refutes it.

Yes, one more precise way to phrase this is that the expected value of the dot product between two random vectors chosen from a vector space tends towards 0 as the dimension tends to infinity (I think the scaling is 1/sqrt(dimension)). But the probability of drawing two truly orthogonal vectors at random (over the reals) is zero - the dot product will be very small but nonzero.

That said, for sparse high dimensional datasets, which aren't proper vector spaces, the probability of being truly orthogonal can be quite high - e.g. if half your vectors have totally disjoint support from the other half then the probability is at least 50-50.

Note that ML/LLM practioners use "approximate orthogonality" anyway.

The Fermi paradox seems to take for granted that complex multicellular life will eventually develop technological civilization, but there is no reason for that to be true. Orcas could very well be more intelligent than humans, but they are physiologically incapable of building tools. Human hands and visual processing were an evolutionary fluke that took tens of millions of years of natural selection, long before apes started using these adaptations to build simple tools. It is plausible that "bipeds with hands" will eventually lead to civilization: getting there is the hard part.

Put another way: birds are very intelligent bipeds, but I am not convinced that the dinosaurs would have developed advanced technology if the meteor didn't hit. There has to be an evolutionary path to develop something like hands, which means either sprouting extra arms or compromising a claw / leg / wing. Perhaps that path shows up, perhaps it doesn't - maybe there are fossils of raptor-like dinos with unusually dextrous claws. But without hands, even superintelligent animals cannot reliably make fire, build ropes, sew leather, carve stone, etc etc.

A lot of these runners come from ethnic groups which live in highlands and mountains: extensive aerobic training in lower-O2 environments, then competing at standard elevations, seems to be the most important advantage: https://www.researchgate.net/publication/225064362_Kenyan_an...

There is probably a minor genetic component, microevolution promoting higher hemoglobin/etc than average, similarly to many Tibetans. But childhood conditioning seems to be a more powerful effect.

This reflexive cynicism is anti-science, because what you claim is not what's happening here. The authors have a sensible causal hypothesis, which is supported by an empirical correlation. A lot of the value seems to be the hard work they did collecting years of urban rat data from various different countries.

Individually, sensible causal hypotheses and correlations aren't worth very much. But when they align and the hypotheses are supported by additional evidence, that's usually a good indication of real causality.

I think of TAS as a code competition, almost like demoscene, whereas tool-free speedrunning is more e-athletics.

Pretty gross to think of these people doing code reviews for projects they have no familiarity with - along with of course this bafflingly incompetent tweet: https://xcancel.com/elonmusk/status/1889062581848944961#m Donald Trump said DOGE were "super-geniuses," and according to his standards they probably are.

(The way Trump's election directly led to "retard" being a common pejorative again hasn't been discussed enough. Just awful.)

Part of it is selection bias, Claude is much less general-audiences than ChatGPT. But any lawyers using LLMs in 2025 deserve to be disbarred:

"A Major Law Firm's ChatGPT Fail" https://davidlat.substack.com/p/morgan-and-morgan-order-to-s...

"Lawyer cites six cases made up by ChatGPT" https://arstechnica.com/tech-policy/2023/05/lawyer-cited-6-f...

"AI 'hallucinations' by ChatGPT end up costing B.C. lawyer" https://www.msn.com/en-ca/news/world/ai-hallucinations-creat...

The list goes on and on. Maybe there's a bespoke RAG solution that works...maybe.

> *By using the term AGI here, we aim to communicate clearly, and we do not intend to alter or interpret the definitions and processes that define our relationship with Microsoft. We fully expect to be partnered with Microsoft for the long term. This footnote seems silly, but on the other hand we know some journalists will try to get clicks by writing something silly so here we are pre-empting the silliness…

OpenAI and Microsoft signed a profoundly silly contract with vague terms about AGI! It's petty to snipe about journalists asking silly questions: they are responding to the fact that Sam Altman and Satya Nadella are not serious people, despite all their money and power.

I understand being against this law on practicality / constitutionality grounds. It seems to me that the "converse" law would be more useful and legally appropriate: forbid people from making intentionally deceptive bots that claim to be human or have human-like emotional/social capabilities.

I like Raschka's writing, even if he is considerably more optimistic about this tech than I am. But I think it's inappropriate to claim that models like R1 are "good at deductive or inductive reasoning" when that is demonstrably not true, they are incapable of even the simplest "out-of-distribution" deductive reasoning: https://xcancel.com/JJitsev/status/1883158738661691878

They are certainly capable of doing is a wide variety of computations that simulate reasoning, and maybe that's good enough for your use case. But it is unpredictably brittle unless you spend a lot on o1-pro (and even then...). Raschka has a line about "whether and how an LLM actually 'thinks' is a separate discussion" but this isn't about semantics. R1 clearly sucks at deductive reasoning and you will not understand "reasoning" LLMs if you take DeepSeek's claims at face value.

It seems especially incurious for him to copy-paste the "a-ha moment" from Deepseek's technical report without critically investigating it. DeepSeek's claims are unscientific, without real evidence, and seem focused on hype and investment:

  This moment is not only an "aha moment" for the model but also for the researchers observing its behavior. It underscores the power and beauty of reinforcement learning: rather than explicitly teaching the model on how to solve a problem, we simply provide it with the right incentives, and it autonomously develops advanced problem-solving strategies. 

  The "aha moment" serves as a powerful reminder of the potential of RL to unlock new levels of intelligence in artificial systems, paving the way for more autonomous and adaptive models in the future.

Perhaps it was able to solve that tricky Olympiad problem, but there are an infinite variety of 1st grade math problems it is not able to solve. I doubt it's even reliably able to solve simple variations of that root problem. Maybe it is! But it's frustrating how little skepticism there is about CoT, reasoning traces, etc.