TeMPOraL

114,503 karmajoined 16 anni fa

Just a programmer who likes to write games and hack Lisp code :). Currently earning a living with modern C++.

Contact me at [email protected].

(I mean it. Don't hesitate, I'm always happy to continue interesting conversations after their HN thread disappears off the front page.)

Homepage: http://jacek.zlydach.pl. More: https://keybase.io/temporal_pl.

I do software nowadays, but I'm constantly looking for opportunities to do something actually useful for humanity. If you know of some, please let me know - especially if they involve biotech, cleantech or NewSpace industry.

(see also: http://jacek.zlydach.pl/blog/2018-01-06-going-far-going-small.html)

comments

4 giorni fadiscuss

Or just AI denialists like to say "LLMs can't do X" even though they can and have been doing it for the past few months or more. They only get called out once the current SOTA LLMs get so good at it, that any rando can trivially and reliably falsify the claim on the spot with whatever SOTA LLM surface they have handy.

Which I suspect is what happened here, given the trail of smaller / local models that successfully answer the question, too.

That said, "curse of reversability" is real, as much for LLMs as it is for people.

5 giorni fadiscuss

> I regularly use a website in which a submit button does not change state in any way. It is indistinguishable from the click having gone to /dev/null. And the completion of the action takes a copule of seconds.

In the good old days, you'd have the request sent by that button reveal a small, transient status bar showing the target URL at the bottom, plus the "Refresh" button in the browser toolbar would change to "Stop" instead; both effects lasted for the duration of the network roundtrip. It was enough of an indicator, and since it was not controlled by the website, it wouldn't break or lie to you.

6 giorni fadiscuss

> The only payoff is convenience. That's how little people and even businesses value their privacy.

Well that, and after 2+ decades of this, we can pretty much conclusively tell it worked out great for them. They were - and are - absolutely right to "make such trades".

Yes, data leaks sometimes happen. Sometimes they even make noise in the news. And... that's about the end of it. There are no tomb raiders stealing "crown jewels" and "secret sauces" and outcompeting companies on their own turf[0] - instead, there are many success stories of systems, products and businesses that wouldn't be created if not for the ability to outsource data and document processing to cloud services.

--

[0] - Except the Chinese, but that's not really about stealing secrets or private data - just that owning the factories lets you iterate on hardware faster (+ it helps to have some healthy disregard for "intellectual property").

6 giorni fadiscuss

Yes, but they'll be doing it between age of 2 and 7, not 20 and 90.

9 giorni fadiscuss

> Flat disagree with this. Small org CEOs are close to their customers and employees and if they behave like dicks then they get punished quickly. Obviously some still do, because people, but it's harder for a small company CEO to continue being a dick.

I'm thinking it might be both - depending on who the real customers are.

I've seen plenty of what I described in "boring" B2C like... grocery stores. But thinking about it, for a grocery store chain, customers are as much a commodity as the products they buy. Suppliers are where relationships (and power plays) matter.

(This might be fundamentally the same problem as the infamous case of "enterprise software procurement" - people using the software aren't the ones paying for it. For a grocery store chain, customers come and go all the time for many reasons, so it averages out anyway - but your suppliers and partners are what makes a difference in your bottom line.)

> And all commercial airplanes, MRIs, anything, were built first by small organisations, and only later by large orgs. Large orgs just can't invent new things unless they form specialist small orgs to do it (skunkworks, or Palo Alto, or similar). Large orgs just don't work like that.

Which is why I tried to point out the category error. "Large org with skunkworks" vs. "Bunch of smaller orgs forming an alliance and acquiring more smaller orgs to productionize a new technology" vs. "government megaproject" - they're all similar, arguably for a given invention they may very well be the same thing. Names and legal groupings are different, but the dynamics is (by anthropic principle) specific to what's needed for a given type of invention.

E.g. for stuff like airplanes or MRIs, you need individuals and small teams with lots of freedom (and a "hold my beer and watch this" culture often helps), but that gives you a prototype at best - scaling this so it works reliably, and then optimizing so it can be economical, both require throwing money at people doing boring work that mostly increments things on margin. And then the money has to come from someone, and someone must be willing to spend it to fund it all.

The actual org charts and legal charters don't matter - what matters is the incentives inside. I somewhat tentatively put forth a hypothesis: large orgs form to solve problems that the regular free market dynamics can't handle, by creating areas governed by different rule sets, within which that work can be done. Whether that's by fiat or corporate charter or a bunch of friends aligning their small businesses for the same goal, is window dressing.

9 giorni fadiscuss

> Do you want to make games, or do 3D engine programming?

Yup. If you start making an engine, you probably won't make a game - especially if you're learning along the way. It's technically possible to succeed at both, but having gone through this process many years ago, and having watched dozens of others in our Polish hobbyist gamedev community do the same, chances are under 5%.

"I'll do an engine for my game first, so it's easier to make the next game" - it's a surprisingly strong trap, because you are actually learning important things and winning small victories every day. There's always another win around the corner. Just one more unroll so the scene looks smoothly. Just one more logic layer to the config format, so it's easier to make the scene. Just one more SIGGRAPH paper.

There's always something important to improve. None of that adds up to a complete game. In retrospect, I'd say, writing your own engine is a perfect way for technical people to procrastinate on the hard but necessary parts of making their dream game - "making it fun", as you mentioned. You end up mastering all the skills that add up to coding an impressive video game. You never learn how to make a game.

(Subtrap of that: you learn a hundred ways to make beautiful visual effects that run smooth in real-time. You never learn how to use that to make art.)

--

(It just occurred to me this is very similar to the trap of "Enterprise Java" - except more insidious, because you're working with concrete terms. Your Scene Graph has no Factory Factory Interface, so you think you've dodged that bullet, missing that it's just unnecessary for the task of putting a bitmap on the screen and making it react to keys.)

10 giorni fadiscuss

And I disagree with yours :).

Large organizations are necessary if you want things like airplanes and rockets and computers and MRI machines to exist. And if you feel you benefit from those things yourself, then large organizations that create and operate[0] them are beneficial to you, too.

> A lot of those products are then enshittified and badly managed because large organisation politics screws things up.

That's not caused by org size. It's how modern economics work because of ad-backed business models and few other things (a tangent for another time). Importantly, small orgs and especially startups are very much complicit in this - the venture capital business model in software settled around a symbiosis, where startups create toys (er, MVPs) and growth-hack the shit out of them, in hopes of winning an acquisition or IPO lottery (aka. "exit"), where a big org buys the whole thing for ${a lot}, and enshittifies it further in an attempt of extracting a positive multiple of ${a lot} from the market. Both sides know what they're doing, exits are planned from day 1, and at no point in this process "creating useful products" is ever a driving goal.

Note this symbiosis: it's a recurring theme.

> Large organisations are inefficient

In some ways. Small organizations are inefficient in others. More at the end.

> (everyone has stories of people in large organisations literally doing nothing all day).

Some (not all) cases of this are about maintaining slack in the system, which is necessary for efficiency. A system at 100% capacity is extremely fragile to breaking completely due to tiny, random workload spikes. Breakage is inefficient. Some degree of idle capacity improves overall efficiency.

> They mistreat their customers and their employees. Their executives tend to lose touch with reality, surround themselves with yes-folk and descend into authoritarian psychopathy.

That description fits small business owners much better IMO. In our times, at least in non-failed western countries, there's a limit to how abusive or careless a large organization can be with their customers or employees - their very size makes them easy to target legally. It might be hard to get through their well-funded legal defense, unless the case is slam dunk, but that's still much better than the armies of small businesses flying completely under the radar, flagrantly violating basic health and safety regulations, or flat out lying to customers in their face, because they're not worth the effort of investigating.

(Of course I'm using a biased sample; I don't know many CEOs of big orgs.)

Symbiosis angle: for abusive practices they can't get away with on their own, big organizations are more than happy to outsource to small orgs and then look the other way.

--

Anyway, key point: *there is no categorical difference between "large organizations" and "small organizations". You need a certain amount of people and communication (and capital) to do high-complexity endeavors. The difference between a well-integrated big corporation, and a hundred of small businesses that kinda end up together delivering something big, is just that the latter is using the market as management layer.

And yes, you need big orgs to create things like commercial airplanes and MRIs, simply because the big org is a boundary layer, within which you have a non-market based incentive structure, and this lets you build things the free market just cannot reach on its own.

--

[0] - Airports and hospitals are themselves large organizations.

10 giorni fadiscuss

My machine has compartments arranged like this:

    [ 2  |  3  ]
     ----------
    [    1     ]

1 is for powder detergent, 2 is for liquid detergent, 3 is for softeners and such.

All three LLMs (Gemini twice, since NotebookLM) insisted I should put the powder detergent into leftmost compartment (2 on the ASCII diagram above). They referred to it by different numbers, but all gave some convincing justification why to put the detergent there. That's despite me posting photos of the compartment drawer, with symbols clearly visible. That's despite demanding they find the manual and cross-ref. I even asked two (Gemini and Claude) to label the actual compartment on the photos I took[0], and both produced some nonsense, with labels in all the wrong places. And they all insisted they're right and issued plenty of warnings about making sure I get this right or else bad things will happen.

BTW. I ended up posting a screenshot of the diagram in the manual to Claude with a passive aggressive comment. Looking at its "thinking summary" and tool calls now, I think at least Claude didn't process the image correctly and only saw:

  [ 2  |  3  ]
  ------------

as those parts are blue, while the bottom is just in the same color as the entire body/frame of the machine. Maybe the contrast was too low for the models. But it was okay for humans, so it's not excusing much, especially that they all claimed to have found the manual, which had a high-contrast diagram.

(Current experience tells me they probably didn't really check the diagram. I noticed recently that all major models seem to have gotten lazy when it comes to reading sources, and are also more than happy to lie about it.)

--

[0] - A method I often use with Claude when I'm not sure if it's dealing with spatial tasks correctly - I have it produce intermediate artifacts that involve modifying "ground truth" inputs - e.g. placing two map pictures on top to verify it solved the coordinate transform, and/or (like here) drawing labels and boxes on top of original photos. I found such requests to be helpful enough I set it as general rule for Claude now.

[1] - Which normally they'd spot, but for some reasons, they didn't.

10 giorni fadiscuss

> I love the asymmetry.

Much as I hate to defend companies climbing to success and pulling up the ladder afterwards, this asymmetry you note is kind of the whole point a company would want to grow big. Growing an organization has some super-linear costs and generally sucks for most individuals living through it - including the management - but it's still considered worth it, precisely because big entities can do things small entities cannot, and escape the threats from smaller competitors.

It's so basic it's actually part of the reason we exist, and animals of various sizes exist, and generally why evolution didn't stop at single-cellular life.

> They have lost the right to cry foul when they trained their model with "but it's fair use" card. Life works by reaping what you sow. Now they are at the reaping stage.

Yup. Except what they're reaping is insane cashflow and ability to pull stunts like these. We can call out the hypocrisy until our throats run dry, and in ideal fantasy land this would've meant something, but here in the real world, they sow the seeds of success, and now are reaping the right to be hypocritical and continue to get away with it.

11 giorni fadiscuss

4. And we need.... something? to realize that both "common sense" / "multiple laypeople" and peer-reviewed studies are right.

Indeed, as per 2., no one is doing the experiments with rocks of different weight, and sufficient heights to easily measure time of fall. However, people have a lot of everyday experience with feathers, grains, leaves, wood, and rocks, as well as objects of various weight made of metal, paper, plastics. And in everyday experience, the heuristic actually holds out well: lighter stuff falls slower, or gets carried away by the wind.

This "heuristic" is purely empirical. You can't disprove it with peer-reviewed studies, because within its scope, it's literally the most basic, purest form of science: direct observation.

So in 1., the mistake is that of incorrect generalization. "Lighter stuff falls slower" is correct for everyday experience, it's the "therefore, heavy rock falls faster than light rock" is wrong.

Not because it doesn't fall faster, mind you - it does[0] - it's just that everyday experience is dominated by aerodynamic effects, and laypeople sometimes[1] mistakenly assign it to gravity.

Which I guess makes it a great analogy for the LLM story. Turns out everyday experience is actually valid in everyday situations. Generalizing from it is usually badly wrong, even if it sometimes arrives at correct answer for wrong reasons (and at wrong scales).

Generalization is hard.

--

[0] - Surprise. It's actually a heavy idealized particle falls at the same rate as light idealized particle. Actual matter is not an infinitely small point in space, and generates its own gravity field, so the heavy rock will land a tiny bit sooner than the lighter one, because it pulls Earth stronger towards itself - but then only if you drop the test bodies one by one (serially), and not together (in parallel, where the difference cancels out). But then it also turns out the mass canceling out for idealized particles isn't just a mathematical simplification, but a very deep truth about the universe...

[1] - Or don't. The question as phrased is, "does heavy rock fall faster than light rock"? This isn't a "specific physics theory question", it's a "real life" question. Treating a positive answer as belief on gravity is an error made by the asker.

11 giorni fadiscuss

Proof is not binary, it depends on the claim and the constraints you put around it, and the nature of the subject of your claim.

Most of the general LLM discourse in our industry is still closer to "proof of the pudding is in the eating" than to "double-blind studies on large cohorts, p<0.01, effect size is still so small that result is useless in practice"[0].

And we're not talking about curated demos either - most of the contested value can be proven for your own specific cases with little to no expenditure of money and time, at a PoC level (it gets more expensive once you try to operationalize it and find kinks that are hard to iron out).

And that is, the article claims (and I agree), the point of last 6-12 months of tokenmaxxing policies and top-down push - it's putting pressure on people to actually go and do those PoC-s for themselves, because just giving the opportunity and permission turned out insufficient for significant part of the workforce.

--

[0] - Ironically, I remember it was the opposite around the time GPT-4 came out. Back then people talked more about specific claims and demanded measured evidence, because it was hard to get the models to reliably do something interesting. But now that the models can handle bad prompting and can understand you even when you're drunk, suddenly people are denying the general capability of LLMs and asking for randomized control trials.

(For double irony, nowadays one can just ask an LLM for randomized trials; the current SOTA models will happily design you a bespoke eval pipeline if you ask them to.)

11 giorni fadiscuss

FYI: I just had three SOTA LLMs + NotebookLM all fail at the simple task of explaining to me where to put a powder detergent in my particular newly bought washing machine, despite having photos of the machine and ability to find the manual (in case of NotebookLM, it literally had the manual as its only source).

After first failure (Gemini 3.5 Flash + NotebookLM), I run the other two (Opus 4.8 on Extra; GPT 5.5 on High) in parallel, and looking at their thought streams, I gave up and dug up the manual and read half of it, before the LLMs finished coming up with - wait for it - wrong answers.

Super frustrating. Doubly so, given that I use them for comparable tasks pretty often and they usually sail through them flawlessly. But this experience happens every now and then. It's only fair to report it, if only so you don't think I'm just AI boosting all the time.

11 giorni fadiscuss

> PTO

I know GP said "available to general public", but my mind went straight to PTO after reading "One of, if not the most dangerous tools". Less common to see (especially in cities) than saws, but I think larger proportion of people understand they have to be careful near table saws than PTO shafts.

> after looking into the crazy world of farmers and military civil engineers

The whole history of aviation and space exploration is chock full of engineers, physicists and chemists doing crazy levels of experimentation.

That said, my point was different - unlike GP, I ask to consider workshops that had extensive experience with dangers of powered or high mechanical leverage hardware. It's entirely plausible and reasonable for people running those shops to say, "here is the new dangerous power tool, it's obviously pretty useful (ask your friends at $X or $Y if you don't see it), figure out how and where to best work it into our specific workflows, so it makes us most bang for the buck".

12 giorni fadiscuss

AFAIU, the whole deal is that the bogus claims never actually reach the DMCA stage - big platforms implement their moderation policies and copyright claim handling specifically to avoid involving the legal system. It's that intermediate layer that incentivizes automated, bogus claims, as there's effectively zero consequences to them.

12 giorni fadiscuss

> They're not "3-4 trillion dollars in investments over 5 years" useful

Why not? They're a general-purpose technology, in the same category as "software" or "electricity".

> nor "crammed into the throat of every employee on the planet, regardless of their actual job" useful

They're potentially useful for anything that can be fed into computers (VLMs lifted the "that can be expressed as text" limitation, visual and audio tokens are not a separate category to text tokens anymore). That touches every single job people do in some aspects. Even though LLMs can't do physical work for people, they're still able to help with directing it and teaching it.

"Cramming into the throat of every employee on the planet" was already covered by many comments here, and the article itself - it's about forcing the obstinate holdouts to at least try.

> Also, you need a more advanced prompt for Firefox on Android :-p

No I don't; literally copy-pasted it to Tampermonkey on my Firefox on Android just now, and it works there out-of-the-box too.

12 giorni fadiscuss

> What? The basic properties of water are scientifically defined as best we can. There is mathematical proof that 1+1=2. Do you think science starts from nuclear physics or differential equations?

Yes. There's a lot of interesting things science has to say about water, very specific claims that took a lot of effort to discover, precisely formulate, and reproduce.

We're not talking about those. The whole LLM discussion on HN, as well as in the wider industry, is still stuck at the state where a large (or vocal) group of people refuses to believe water is wet. Yes, there is a similar group that tries to sell water as miracle cure, I'm not denying it - IMO both perspectives are dumb and entirely detached from obvious observational evidence that you can collect for ~free at home in 15 minutes. Example will follow.

There exist the equivalent of foundational, detailed studies on LLMs, at every level of rigor imaginable (with a caveat, it's hard to rigorously prove anything useful in software engineering; it's still largely opinion-driven field). But they're not part of the overall "AI hypers/haters" dynamics.

> I wish HN had a way to classify you as an "AI booster" or equivalent.

You can take any of the LLMs and have it vibecode you a user script in under 5 minutes, than you then can paste into Greasemonkey/Tampermonkey, and voilà, you have me labeled as "AI booster" or filtered out.

In fact, let me help you, I'll time it. I opened chatgpt.com in incognito (to emulate being a rando free user), and put the following prompt in:

> I need a user script I can paste into Tampermonkey on my Firefox that will clearly label user named TeMPOraL with robot emoji and some silly emoji, so I never forget when reading their HackerNews comments that they're an unapologetic AI booster.

Got back this script in under 10 seconds: https://pastebin.com/akEchvHd. Tested it, works out of the box.

This is the promised empirical example. It doesn't prove everything, but it proves something, and it took, end-to-end, a total of 1 minute to perform just now. You can collect many such examples over a single day by just trying. People who keep saying AI is useless and a fad and can't do anything useful, obviously never bother with even that.

FYI: I'm not an AI booster. I like AI, and I find it useful, but I'm not going out of my way to boost it. I just enjoy this topic, but more importantly - and I remain consistent in this - I point out bullshit that doesn't agree with obvious observable reality.

EDIT: try the example yourself, and post whether it works for you too - if it does, it's technically a peer-reviewed, replicated study, but I doubt it'll convince any of the naysayers of anything.

EDIT2: I have plenty of negative things to say about LLM capabilities and how irresponsibly people use them, and I do occasionally write about this (mostly at work, these days), but most HN threads on AI are not on this level - not anymore. They used to be more reasonable back in GPT-4 days.

12 giorni fadiscuss

If having a job is still marginally better than not having it, it's a sign of an efficient labor market.

(Market inefficiencies is where everything that is nice, beautiful, good, human, lives.)

12 giorni fadiscuss

Source: it takes less effort to test this yourself than to write comments about it on Hacker News.

12 giorni fadiscuss

LLMs are DEI-aware, as over past few years, their vendors all had various high profile news stories with their models and their default biases, so it's more likely they'll heavily discriminate in favor of minority candidates, not against them. Still, in both cases it would indicate whoever is operating the system is doing a really, really lazy job. It's really not hard to test and supervise LLMs on tasks where they give you mere 2-10x leverage, and prompt adherence today is much better than it was 3 years ago.

12 giorni fadiscuss

Nature doesn't publish papers about water being wet either.