cornholio

With what accuracy? And with how many intermediate tokens?

So you are arguing here that LLMs should just "know" the result of a multiplication when the operands are in context, ie, that a hidden multiplier circuit should emerge in their weights.

Is this how you do multiplication, if I give you two 12 digit numbers, does the 24 digit multiplication result just pop in your head? Don't you have to follow a learned algorithm through a tedious system 2 effort? Don't you need to write down the results on paper because you can't actually hold in your head the dozen partial results, each with a dozen digits? How many visual, tactile and reasoning tokens does this consume, moving the hundreds of muscles that make up your hand to draw each number under visual feedback, then reading all those numbers back and transforming ocular activation data into numeric symbols?

It seems to me your system 2 is just following a symbolic algorithm for multiplication, and it does roughly the same steps as the LLM trace I showed you previously.

So if you think this is the mark of your intelligence, why wouldn't it apply to the machine too? Why is it implausible that, following a similar algorithm learned from some mathematical paper, the LLMs has reasoned a new solution to a problem in another math field? Why couldn't the machine combine and morph these algorithms for symbolic manipulation, to yield entirely new and original results? How would those results differ from results human mathematicians generate, using recipes they learned in university?

The emergent behavior that we talk about isn't that the machine can do multiplication in its "head" after knowing the multiplication algorithm. What emerges is the ability to follow any other algorithm, even algorithms that were not in the training set, even algorithms to create other algorithms, which it then executes. This is the emergent behavior that matters for AGI; once it can do that, it's a trivial exercise to create a non-AI tool to automate and accelerate the mechanical tasks - just like we humans do it.

The field now favors into the view that symbolic manipulation is not the mechanism of general intelligence, but rather an emergent byproduct of learning. So the fact that a connectionist machine (neural network) got so good at symbolic manipulation actually supports the view that we are closing the gap to general intelligence. Through the rote work, the machine really internalizes those rules and the symbolic manipulation capabilities are emergent, just like we humans do it.

What still confuses people is the insane inefficiency of deep learning, and that those emergent capabilities require such an immense training corpus compared to the only other architecture that we know of.

But this already is an optimization problem. If the machine gets super human at symbolic reasoning, and at the same time, can solve the symbol grounding problem to real world data and sensors, what prevents you from saying it thinks? Can it not solve real world problems? Can it not redefine its tasks and display some form moral agency - even if a totally foreign morality for us humans? Can it not use these abilities to reproduce and expand, create ships and turn the universe into paperclips, if it finds it worthwhile?

Math is basically just a playground that is perfectly suited for these emergent capabilities, so of course we will see the first progress here; but there is no firewall separating math problems from general cognition.

I think it's becoming harder and harder to argue that LLMs don't really reason and just mimicry human speech. This counterexample is clearly the result of a sequence of steps that build on previous knowledge in context and logically combine it to reach other true statements - to a degree and complexity that rivals the best human minds.

For someone that use Claude Code every day, this is obvious, but for some reason many scientists refuse to accept that it's truly reasoning; perhaps not in the human sense, but in a very profound and real sense. These powerful results are devastating to their point of view.

I can sympathize, because I too called LLMs "fancy Markov chains" in the GPT 3 era. But there comes a time where you have to update your world view to match reality, or be stranded in fantasy land.

Arguably, Exxon's plan to build a computing ecosystem rival to IBM's would have worked too, if the 16 bit successor of the Z80 would have maintained upward binary compatibility with Z80.

CP/M was an absolute beast in the era, with massive installed base and software support, employing 500 people in 1982. A CPU that could run unmodified Z80 software in a 64k segment would have allowed DRI to ship 16 bit CP/M with only basic tweaks and likely kill the market for the PC.

It was, famously, DRI dragging their feet on 8086 support that motivated the release of QDOS, which was then bought by Microsoft and relicensed at an immense markup to IBM as MS-DOS.

Assuming the necessary laws banning this crap are not put in place, what is the endgame here?

Activists destroy visible surveillance cameras, so they hide them and make them hard to recognize. Activists trace the camera locations from the public data, so Flock kills those feeds and sells only to vetted buyers.

The value of mass surveillance is high enough and the power imbalance so strongly against the citizenry, that someone will setup these hidden cameras, as long as it's legal.

Imagine what you can do with this data, face recognition and GPT-5 class agents. Not only do you have the realtime location of your victims, but now you can see who they talk to, what they wear, what mood they are in, what they bought, are they drinking or visiting a brothel, what car they go into - and it's no longer an ephemeral cookie id, it's the face that person will have forever, on their id documents, in any interview or loan application they will ever do.

This data is worth trillions in the long run if sufficiently oppressive structures are put in place to leverage it.

In a society that respects and protects privacy rights by law, the service Flock provides could not exist.

There are no 'guardrails' to mass surveillance.

This is another reframing of the problem. What makes the area desirable is previous investment society has made there, to the benefit of some and the exclusion of others; the scarcity is created by political choices, not intrinsic.

A baseline of housing - for example, a Japanese style capsule or ultra-tiny home, that you can lock and store your belongings safely - costs close nothing. Of course, nobody would want the hobo-hotel in their neighborhood, but it has nothing to with scarcity.

the limited availability of land zoned for housing.

A limited area of land is zoned for housing because those with the power to expand it are already housed. This explains how scarcity is created, not that there is any intrinsic scarcity.

It's circular to say "it's an allocation problem". Yes, that's the entire point: we're post-scarcity on food supply and yet, as a species, we can't guarantee the allocation of a livable baseline to every person.

So, it's reasonable the same "allocation problem" will plague the AI economy: some will "thrive" and get to control the output of the auto-factory, some will get nothing.

They can't, because they use RL with synthetic data and LLMs as judges. So the system naturally convergences towards certain load bearing, genuine, not just annoying but ridiculous verbal tics.

It's probably the reason most LLMs share the same tics across labs, because they cross train and distil each other's models on an industrial scale. You also can't escape it in generated text that's already online. So if, say ChatGPT first had some random idiosyncrasies, it then contaminated the entire AI ecosystem.

It was quite clear 4.7 was a dumbed down high efficiency model they put out in a rush to handle the capacity issues they were having at the time. I've experienced myself substantial degradation on basic reasoning tasks, which were fixed in 4.8.

Can't wait to see what unusually simple Erdos problem LLMs will expose next, hiding in plain sight for decades and seemingly intractable for professional mathematicians who weren't aware just how simple the problem was.

You are still getting the models you signed up for. The all you can eat added a French wines selection - which requires separate payment.

This is true for any drug, any drug can presumably become a poison, can interact with some genetic or biological trait and trigger a side effect and so on. The complexity of the biological systems is so great that they defy clear deterministic understanding, but stochastic empiric knowledge and treatments still have immense value.

If you give me an inference chip that runs 200x faster, yes, it could be backdoored to take control of my dishwasher and kill me in my sleep - but I can't deny it runs 200x faster an account of nobody being able to explain why. The same for the mistery cancer drug that cured everyone who took it up to now, but could, without doubt, kill the next patient.

It's easy to find counterexamples: the entire science of pharmacology is based on macroscopic effects that often lack a fundamental understanding of the underlying mechanisms of action. Psychopharmacology is the extreme example. Often, the fact that a drug worked made scientists investigate and discover the mechanism behind it, but for many drugs used every day by billions it's still a mystery, or it's understood only in very broad terms.

So what will you do if the doctor prescribes you an LLM-vibecoded drug that nobody understands how it works, yet it cures some deadly affliction with close to 100% efficacy?

What if, say, these incomprehensible math results lead to a revolution in quantum physics which unlocks chip topologies that are orders of magnitude faster than human comprehensible designs?

Would the high priestess of human reason pass her divining rod over such chips or life-saving drugs and reject it as the work of the AI devil?

Personally, I would avoid UTF-8 levels of complexity because you only pay the size cost once per string. A simple 2-bytes + optional 4 bytes continuation scheme handles strings up to 140TB and increases the size of the average string by just 2 bytes (compared to 1 byte for nul termination).

You can have a universal variable length field, for example 2 bytes for strings < 32768, then four bytes, 8 bytes etc. On the critical short string path, it costs just a single bit test. The glyph vs byte issues need to be dealt with in both formats.

The subdivision issue is a good perspective, but i would argue the performance impact of cloning substrings is dwarfed by the redundant full string reads to find length.

The point is that knowing what you don't know allows you to hedge your bets on events far into the future; my child will need to become productive 20 years from now and maintain it for 40-50 years. So, it's a bit like trying to educate a kid born in the 60s for the web era, you can't and you shouldn't even try.

What you can do though, is to offer them broad exposure to things that are interesting to them and their generation; my eastern block clone of the 8bit/48KB Spectrum computer didn't really help me excel at math, reading or history, nor was it to be the future of technology, but it did change my life significantly by letting me understand and relate to people that I couldn't otherwise have business dealings decades later.

It seems imprudent to cut children off from futuristic technology just because of a moral panic that it causes brain rot. Unless we know it's soma, a drug so powerful that it subdues volition and curtails intellectual development; we don't.

The counterargument is that kids will live in a world different from our own.

For example, in many countries children lost the ability to write cursive; that used to be a critical skill comparable to literacy itself. But in our current society, that's no longer the case and you can be very successful without it, but there are other skills, such as using technology, that became critical.

Any definitive claim to know what are the right things kids should learn in a moment of rapid technological shift is probably garbage and just a projection of our own biases.

rewriting a big widely used project in a stricter language is always a good thing

Always might be a too strong word. Rust is, by design, a language with low development velocity.

So you risk: 1. ossification of the current architecture and deferment of important features; or 2. reliance on AI coding to recover velocity.

Maybe for some 2 does not look like a risk, but I think it's too early to call. We have yet to see the effects of extensively using these tools on large scale projects, for years and decades.

If radiator temperature is critical you can just run a heat pump to boost it. Modern heat pumps can get pretty close to the Carnot limit for the temperature interval, for example a pump cooling to 300K and dumping out heat at 400K (125°C) will have a theoretical COP limit of 4 and a practical limit of 2.5-3.

That means that for every 3 units of heat your chips emit, you will use a 4th unit to spin the pump. If your panels generate 150kW, you will only have 113kW available for compute and the rest is cooling. Radiators will more than halve vs 340K operation, so it's net beneficial.

It's all a giant techno-economic optimization problem: the extra mass of solar panels you need vs saved radiators vs mass of the pump, the die temperatures you achieve and corresponding performance, durability and chip price point etc.

I'm not sure they can actually respect that 30 days absolute commitment. Let's say some internal tool flags a suspect conversation, it bubbles up and a human operator reads it and it looks like evidence of a crime. Then, that employee is legally bound in many jurisdictions to prevent the destruction of that piece of evidence.

It's one thing to commit to a "everything is deleted when you press delete" automatic policy. It's quite another to say "we'll keep some stuff for up to 30 days, look inside it for any malfeasance, then pinky promise we'll delete it".

I'm not entirely sure how that dismisses the CEO's putative argument: they go big on AI precisely because shipping end-to-end is hard, so they think they shouldn't waste resources on tasks that can be automated.

The structure of a good argument would be something like: certain tasks are fundamentally human and impossible to automate (which and why?) and by pushing AI use beyond what is optimal you are actually hurting your employees ability to do those hard parts.

A weaker but still useful argument is that most everything can probably be automated, but frontier models aren't there yet.

Focusing on pure energy efficiency might be missing the point of economic efficiency.

An RO desalination plant needs electric energy to drive the pumps, which might be generated by panels which are 15-20% efficient. So, if you can have cheap thermal desalination panels, they come out ahead even if 6x less energy eficient, you avoid the whole expensive and fragile desalination plant and you gain a low skill, distributed setup.

The LVCVA is a public authority and the Loop contract had substantial penalties in place, about 30% of the total ~$50 mil amount, tied to various tiers of service the system needed to hit. So right off the bat not only do you have good financial incentives for a good process, but strong legal risk and public scrutiny if things go bad. While the test was not public (for these very reasons), it was attended by auditors from BDO LLP - LVCVA's long standing auditor (and not one hired for the occasion by BoringCo).

The accounting company affirmed the test results, CFO Ed Finger told the board that the auditors observed 157 unique rides and there were no negative findings. These are all public records and board minutes anyone can request and consult.

Essentially, according to what was reported in the media, they had the few hundred volunteers board packed Teslas (3 pax + driver), ride the system, disembark and take another ride, while remaining within the station limits.

I wouldn't call the 4400pphpd result an "extrapolation" - it's a real, instantaneous capacity number once the system reaches steady state, just like you don't have to drive for an entire hour to express your speed in km/h or mi/h.

The figure is of course not indicative for the real rush hour capacity of the station infrastructure, especially the underground stations that have escalators etc., when they are packed with disoriented tourists carrying luggage and not necessarily making an effort to move fast and hit good numbers.

Hence, I think my original point stands firm, Boring has demonstrated vehicle rates that can beat light rail, the basic premise of cheap narrow tunnels is sound. To actually demonstrate competitive numbers in real life scenarios will require larger vehicles, better organized ingress flows and procedures, more and perhaps larger stations etc. But these are all tweakable factors depending on actual demand at a specific station, they can run a mix of vehicles and expand infra where it makes financial sense, but only IF the tunnels have good vehicle rates, ie. no more than seconds of headway with extraordinarily rare in-tunnel break downs.

Hitting almost 4000 was a requirement of the LVCC bid and they actually hit 4400.

This is all public info available from unbiased sources; preferring schadenfreude and rage inducing bait is, of course, a choice you are entitled to.

Loop demonstrated 4500 people per hour with Teslas with human drivers. Hence, the assertion that it can be competitive with most light rail systems is entirely reasonable, assuming, as I said, "they get adequate, automated, larger vehicles". For example, the much hyped but so elusive Robovans.

The largest subways in the world can reach 80,000 pphpd (crush load) but the vast majority of US systems are under 20k, and those are numbers Loop can likely reach with larger vehicles:

https://www.reddit.com/r/BoringCompany/comments/s4r6l4/chart...

Only a handful of the largest US systems really hit the capacities the only heavy rail subways can support, and they do so with eye watering costs, see the famous $2.5 billion mile in New York's Second Avenue extension. So a $10-20mil/mile system with 1/4-1/2 the capacity of a full subway could, if demonstrated, completely change the game in many cities.

It's interesting how Loop criticism has evolved over the years: initially, it was claimed Boring Co has no chance of duplicating or significantly improving on the speed and cost of state of the art used by tunneling companies, or that the entire concept is fundamentally impossible or cost-prohibitive to build. After building the first stations, it changed to capacity concerns or that it's not Hyperloop, an completely unrelated high speed concept.

Now, after Loop demonstrated vehicle rates that would beat most light rail projects and many subway systems too - assuming they get adequate automated larger vehicles - the criticism seems to be that it's not build out yet to a significant size, or that it has minor usability quirks.

I sure hope these people understand just how foolish that position risks to become if Boring Co continues good execution and build outs the entire system as planed.

the great problem of our age is deciding who controls technology

Isn't that just an instance of the political problem for all ages: who controls what, who gets to rule and who obeys, the fundamental power struggle apparent in all human history.

Extend the definition of technology to the broadest sense, from the material that allow us control over the physical reality: steam, computing; to the organizational, that enable collective human action: states, factories and assembly lines; and the ideological, that legitimize certain power arrangements: religion, nationalism, democracy, human rights etc.

A feudal lord's power rested on land (material), the manorial system (organizational), and the divine right of kings or religious sanction (ideological). Even if peasant revolts happen from time to time, the arrangement is stable because the peasantry accept it as legitimate and have no economic alternative; so even when revolting they cannot imagine a different political order. Technological (broad) leaps like the Enlightenment and the Industrial Revolution change the political possibility space so arrangements like feudalism are no longer stable, but others like capitalism, liberal democracy etc. become possible.

Political actors observe these technological shifts and struggle for control, relevance and power. The old elites are contested by the new kids on the block, wielding the new technologies: the aristocrat by the bourgeois, pastoralist tribes by agricultural states, autocrats stuck with traditional propaganda by the kids with smartphones and social media.

The present struggle around AI is therefore to be expected; what's more interesting is the type of political possibility space it opens up: is it one where having the bulk of society educated and productive, capable of running the machines is the key factor pushing the country forward in the international technological competition, like we've see post-war, forcing the national elites to cater to their needs, invest in their populations and broadly share the economic output and the political power? Or is it more likely one where the key competitive factor is the size of your datacenters and automated defense factories, where the bulk of people are irrelevant for the architecture of power?

Because if it's the latter, the entire idea that democracy will somehow manage to survive and influence who gets what becomes problematic. In the new technological-historical space, democracy becomes structurally unfavorable and thus, unlikely to persist long term.

That's a JIT. Yes, you can do all sorts of optimizations in a JIT, because you do it at runtime using runtime information, and always keeping an escape hatch, so the static code bails when invoked with data it was not compiled to handle. This kind of hatch is used here with <any> wrapping.

JIT is a technique to accelerate dynamic languages at runtime to near machine performance while keeping dynamic ergonomics; but it can't transcend the AOT / runtime wall.