computably

Sure, if you're shipping a 500MB mobile/web app with a 5s start-up, it probably won't make a difference today, or tomorrow, or 3 months later. The problem is when your codebase is a dumping ground for AI slop and a year from now nobody can make any changes.

Also, it's not just some subjective value put in human review. If a change causes a major outage and it turns out it's because 5 reviewers were all shirking the company-mandated responsibility of review, and it could trivially have been caught, I would 100% expect those people to get fired. It's not a process issue if people are intentionally circumventing the process.

A term that gets tossed around in math is "mathematical maturity." It's similar to what you see in other fields - e.g. learning how to program, learning how to make music, learning how to cook - that involves many "aha" moments and reshapes your perspective. Math is full of such steps, moreso than most other endeavors, probably because the main limit is the abstract reasoning itself.

The asymptote is log2(3) = 1.58496...

That's a backwards argument. Flights in California between SF and LA will obviously travel a continuous line across California. Yes, the populated areas have light pollution, but that's because they're, y'know, populated areas. HSR itself would add nearly no light pollution outside of the existing population centers, in contrast to flights.

Lol. The average American isn't ponying up $20/mo for the latest and greatest model. Barring major breakthroughs, being months ahead is a race to the bottom of who can burn more money to get 0.1% on a benchmark with no practical relevance.

Except, of course, LLMs are not humans, and they do not learn or "reason" in a way which even remotely resembles humans.

Plus obviously humans can still overfit to a specific style of test.

> Are there examples of where we have collective decided not to pursue knowledge? Successfully?

Intrinsically, the knowledge humans choose not to pursue will not be much publicized. There's limited value in calling attention to it and it doesn't make for good entertainment. Plenty of examples provided by other comments nonetheless.

> Perhaps I'm horribly naive, perhaps I just see the SciFi future I've spent my life reading and dreaming about on the horizon and I'm blinded by the reality, perhaps my ideals around "knowledge deserves to be free/accessible" are misguided. I don't know.

I don't personally think there's intrinsic benefit in disseminating arbitrary knowledge. There's quite some difference between the printing press and nukes.

Applying the term NIMBYism to anti-AI and anti-DC sentiment is a gross abuse of terminology. Datacenters don't need to be in anybody's residential neighborhood.

LLM output is non-deterministic. Even if the attack fails 50% or even 99.9% of the time, at YT's scale it's a pretty huge issue.

> Naively, this probably takes a lot more RAM (not to mention all the other computing resources) than to just have another HTTP handler sitting idle on some computer somewhere.

CF Workers uses V8 isolates rather than (OS/HW-level) VMs.

Also, the RAM usage isn't really a differentiator between FaaS and any other cloud instance. If a VM is used for FaaS, it's almost certainly also used across the platform for multi-tenancy, invisible live migration, etc.

Inserters have a max throughput of 120/s each chest-to-chest, 12 inserters can (un)load a single cargo wagon so 1440/s per-wagon. So 6:1 belts:wagons, a massive speed advantage over long distances, and much more flexible M:N distribution.

Huh? I thought a stacked belt holds 4x60=240/s

edit: It does seem to be 240/s: https://wiki.factorio.com/Transport_belt_capacity_(research)

I feel your comment is deeply ironic... Just shy of self-awareness. Have you considered that the conceit of the fictional character is representing that reality?

> It applies its knowledge to create bespoke solutions to the problem you pose to it, and is able to self evaluate its progress towards the completion criteria.

It imitates applying knowledge. The imitation may be uncanny, but assigning LLMs intentionality and ToM is a category error.

Presumably they meant that they'd sacrifice some material value for some animals, not that every animal on Earth has infinitely more value than inanimate goods.

> infinities cannot be compared

That's either a mathematically illiterate assumption or a very strange philosophical hill to die on.

> some tragedies cannot be averted

Sure. The question is what to do about the ones that can be averted.

> some decisions are impossible to make

> and "prioritization" is a distraction that forces choices when choices are not strictly necessary.

Again, the question is what choices to make when you can (arguably must) make them. Saying they're impossible is just refusing to take responsibility. You either do something, or you don't.

> but this is the first time I’ve experienced software that feels like it’s actively trying to be disrespectful

It sounds like they use plenty of software so they must be incredibly lucky, picky, or both.

Except LLMs are simulacra of actual intelligence. Frequently in a single conversation working on a single narrowly scoped task, I am both surprised by a few insights and cursing at how it can miss obvious issues. The "raw intelligence" of LLMs leaves much to be desired.

> If someone will sell you a service for a dollar bill or three quarters, why wouldn't you take the three quarters?

Because one day they'll send you an email informing you the new rate is $1.50, and if you missed the email, that's not their problem.

Storage is multiple orders of magnitude slower than RAM. Pretty sure it'd be more like 10s/tok than anything reasonable.

Akshually, they said "harness," and not "test harness."

There's no particular reason "agent harness" can't have practically the same definition, substituting test-specific concepts for agent-specific ones.

> can't you just let people enjoy things?

Dumping slop into the public commons deserves criticism.

"Best" is subjective. But "caring about their users"? Their response to RtR alone shows they care about their margins more than their users.

> We fit continuous theories to discrete measurements--and the good ones fit really well!--but until we can measure it how can we actually know?

Well, physicists came up with quantum mechanics because they found a way to distinguish a genuinely discrete phenomenon.

Understanding the physical universe overlaps with a subset of math. It shouldn't constrain the abstract tools which may or may not one day be useful for that understanding.

> An instance of a number that has a special meaning.

Not really. There are infinitely many infinities. Infinite numbers are not particularly more special than real numbers, complex numbers, matrices, functions/operators, etc.

> It's deeply distressing to watch people fall into AI psychosis.

It's unclear what you're saying here... Yes, AI-induced psychosis is a real problem and the frontier labs' mitigations are ineffective, to put it mildly. But using AI as a coding tool doesn't have anything to do with psychosis.

> I suppose they'd be comparably successful.

Yes, so, not particularly.

A strange view. The trade-off has nothing to do with a specific ideology or notable selfishness. It is an intrinsic limitation of the algorithms, which anybody could reasonably learn about.

Sure, the exact choice on the trade-off, changing that choice, and having a pretty product-breaking bug as a result, are much more opaque. But I was responding to somebody who was surprised there's any trade-off at all. Computers don't give you infinite resources, whether or not they're "servers," "in the cloud," or "AI."

> Also by the way, caching does not make LLM inference linear. It's still quadratic, but the constant in front of the quadratic term becomes a lot smaller.

Touché. Still, to a reasonable approximation, caching makes the dominant term linear, or equiv, linearly scales the expensive bits.

> That's super interesting, isn't Deepseek in China banned from using Anthropic models? Yet here they're comparing it in terms of internal employee testing.

I don't see why Deepseek would care to respect Anthropic's ToS, even if just to pretend. It's not like Anthropic could file and win a lawsuit in China, nor would the US likely ban Deepseek. And even if the US gov would've considered it, Anthropic is on their shitlist.

> It seems basically impossible for everyone to have overhired, for the simple reason that qualified workers do not appear and disappear from nowhere. There is a population of qualified workers in the software sector, and only new grads and retirement can move the needle significantly.

SWEs (and most any role for that matter) definitely can be minted in ways besides graduating with a relevant major. On top of that there's also H1Bs and contractors. Plus "overhiring" doesn't necessarily just mean absolute headcount, it could be compensation, scope, middle managers, etc. The definition of "qualified" is also malleable depending on the incentives.

> So, if someone overhired then someone else must have done without, all things considered.

Beyond the previous points, this also assumes the supply of labor is independent of the demand, and it's clearly not. As the demand increases, so does compensation, outreach, advertising/propaganda, etc. Everybody can overhire simultaneously as a result of pushing for growth of the supply of labor.