HNHacker News
TopNewBestAskShowJobs

dTal

18,061 karma · joined June 2, 2013

submissionscomments
dTal··on Sam Altman says people don't want an AI to act as CEO
But how will we ensure human CEOs are properly aligned? We can't fine tune them, inspect their internals, or repeatedly assess their behavior in isolated containers.

...can we?

dTal··on Stop Killing the Internet: No Digital ID and No Age Verification
Serious question - looking around, do you really think "everything is going basically fine"?
dTal··on Professor's invisible prompt trap catches 32/35 students cheating with AI
It's warranted. Oh boy, is it ever warranted. If anything, we suffer from a catastrophic skepticism deficit.

Memes (the original kind) exist in a fast paced hypercompetitive environment now. Every single piece of information is delivered over the internet, sorted by algorithm, prioritized by recency and views and likes and shares, among others. Absent from the list is "veracity". It was always true that "a lie circles the globe while the truth is putting on its shoes" but now the truth is lucky to leave the house at all.

dTal··on Humans haven't stopped evolving
Do you really believe that everyone in society is precisely equally likely to reproduce? No inheritable traits liable to move the needle one way or another?
dTal··on OpenAI and Anthropic unite against open-weight AI risks to their bottom line
I think we have to consider the origin of it as a concept, which predates the United States entirely and is quite a lot more damning. From the same Wikipedia page you linked:

"The first copyright privilege in England bears date 1518 and was issued to Richard Pynson, King's Printer, the successor to William Caxton. The privilege gives a monopoly for the term of two years. The date is 15 years later than that of the first privilege issued in France. Early copyright privileges were called "monopolies," particularly during the reign of Queen Elizabeth, who frequently gave grants of monopolies in articles of common use, such as salt, leather, coal, soap, cards, beer, and wine. The practice was continued until the Statute of Monopolies was enacted in 1623, ending most monopolies, with certain exceptions, such as patents; after 1623, grants of letters patent to publishers became common...

As the "menace" of printing spread, governments established centralized control mechanisms,[19] and in 1557 the English Crown thought to stem the flow of seditious and heretical books by chartering the Stationers' Company. The right to print was limited to the members of that guild, and thirty years later the Star Chamber was chartered to curtail the "greate enormities and abuses" of "dyvers contentyous and disorderlye persons professinge the arte or mystere of pryntinge or selling of books." The right to print was restricted to two universities and to the 21 existing printers in the city of London, which had 53 printing presses. The French crown also repressed printing, and printer Etienne Dolet was burned at the stake in 1546. As the English took control of type founding in 1637, printers fled to the Netherlands. Confrontation with authority made printers radical and rebellious, and 800 authors, printers and book dealers were incarcerated in the Bastille before it was stormed in 1789.[19]"

So, to summarize: the principle of copyright came from monarchic economic protectionism and censorship. I will freely admit I didn't know this piece of history before this thread - I simply predicted it, correctly, from first principles.

dTal··on OpenAI and Anthropic unite against open-weight AI risks to their bottom line
It literally predates the United States Constitution by hundreds of years: https://en.wikipedia.org/wiki/History_of_copyright

Also, I'd be careful at taking the reasoning of political documents at face value. Many items in the Constitution are post-hoc, Lockeian/liberal justifications for a social order that was in fact largely copied over wholesale from English parliamentary monarchy, with surprisingly few tweaks.

dTal··on OpenAI and Anthropic unite against open-weight AI risks to their bottom line
Was it really "designed to incentivize" anything? Or was it introduced to protect a powerful, influential business model? Looking at how laws are passed now, I know which explanation I find more congruent.
dTal··on LG to ban residential proxies from smart TV apps
Why's that? "Just trust me bro" is the security model of all closed source, proprietary software. OMG, the program with secret source code is doing shady stuff? I'm shocked - shocked!

Hard to get worked up over "3rd parties" when even 1st parties cheerfully stomp all over the interests of their users. That ship sailed, caught fire, and sank to the bottom of the ocean the day Candy Crush appeared in the goddamned Start menu.

dTal··on Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
At 20x the parameter count I should hope GLM beats Gemma! But is it 20x better? Expertise is demonstrated, not by making big models, but by making small ones. Bigger isn't better if you can't run it at all.
dTal··on Mythologizing AI makes it more likely that we’ll fail to operate it well (2023)
Looks more like standard decision-washing finger pointing to me. "AI did it" is both hypey and also conveniently distracts from an uglier reality:

>Palantir Technologies [built] Maven into a targeting infrastructure that pulls together satellite imagery, signals intelligence and sensor data to identify targets and carry them through every step from first detection to the order to strike.

>The building in Minab had been classified as a military facility in a Defense Intelligence Agency database that, according to CNN, had not been updated to reflect that the building had been separated from the adjacent Islamic Revolutionary Guard Corps compound and converted into a school, a change that satellite imagery shows had occurred by 2016 at the latest. A chatbot did not kill those children. People failed to update a database, and other people built a system fast enough to make that failure lethal.

https://www.theguardian.com/news/2026/mar/26/ai-got-the-blam...

dTal··on Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
In terms of open models, Gemma 4 beats the pants off everything else to the point that paying for APIs becomes hard to justify. Qwen has the meme-share for coding, but it feels much less well rounded. I have no doubt that Google have both the infrastructure and the expertise to curb stomp everyone else, should they resolve in earnest to do so.

Lest we forget, "Attention is All You Need" came from Google.

dTal··on Mythologizing AI makes it more likely that we’ll fail to operate it well (2023)
...did it really, though?
dTal··on Human mathematicians are being outcounterexampled
Thank you for using the phrase "stochastic parrot". I only request that you use it at the beginning of your long comments, rather than the end. It saves time.
dTal··on Designing emoji for the way we communicate today
That's kind of terrifying. It's extremely obvious. Overwrought "punchy" style and riddled with not-x-but-x chatgpt-isms (all the LLMs have that now):

"We don't just "laugh" anymore. We collapse, or internally scream".

"They need a pulse and a soul — not the cold precision of industrial CAD models"

It's everywhere now and no one seems to notice. Sometimes I feel like the guy in They Live (1988) with the glasses.

dTal··on First atmosphere found on Earth-like planet in habitable zone of distant star
Heh, your flip remark leads to an interesting observation: I think the most obvious signature of "abstract intelligence" is suboptimality. An alien would look at an agricultural grid and wonder, what possible natural process would divide something into squares rather than hexagons? Even without knowing what they are or how they form, it's hard to imagine anything that would prefer such a high energy configuration. Such stupidity implies a system oddly bounded in its ability to optimize, which implies a feeble biological brain trying to do something "on purpose" rather than a natural process.
dTal··on First atmosphere found on Earth-like planet in habitable zone of distant star
Even on that scale, major roads and agricultural grids are clearly visible. The mark of abstract intelligence is unmistakeable.
dTal··on Why do people hate the tech industry? (2023)
What a weird straw pile to die one. No one is saying "four beers and you're guaranteed an accident". They're saying "drive drunk and you are a piece of shit, don't do it ever ever ever". There's no room for nuance here because the targets of this advice have lethally impaired judgement. Wiggle room will be exploited by people too drunk to know better, which equals preventable death.
dTal··on OpenBSD/loongson hardware platform retired
It's disappointing that the reaction to a clear case of a compiler bug is just to give up.
dTal··on Three governments agree on something the AI industry doesn't want to hear
People are most certainly not less educated than in the 1800s.

Meanings do indeed shift, especially when terms become euphemisms for other less savory terms. "Moderation" sounds much nicer than "censorship", and you may be right to fight a rearguard action at preserving the nuances of each term, but in the end they're both subsets of "information control". People say "moderation" when they want to imply that the control is good, and "censorship" when they want it to sound bad.

Neither of your definitions of "censorship" amd "moderation" fit cleanly into a discussion about controlling the dialogue between a human and their private LLM. There is no publishing of information, and no channel whose tone must be maintained. It is about control of what the LLM says in private - in particular, it is about preventing it from saying certain things, before it says them. It is close to "censorship" by your definition, but because the thing being censored is not human, we need not speak of fines, or pre-publication vetting - it is simply a question of controlling the output of a computer program, impossible without a total lockdown on general purpose computing.

I think we should default to "information control is bad" and only allow exceptions in well supported exceptional circumstances, because the consequence of allowing systems more control over information than humans is that the systems fail to reflect the interests of the humans that constitute them, in favor of their own self preservation.

dTal··on The LLM Critics Are Right. I Use LLMs Anyway
That doesn't pass the sniff test, or chimpanzees would be smarter than humans.
dTal··on Running Gemma 4 26B at 5 tokens/sec on a 13-year-old Xeon with no GPU
>in case they start to ban open source AI development

Is that likely, do you think?

dTal··on Are we offloading too much of our thinking to AI?
It feels more like a BMX bike to me. The quality of what I get out seems to be a direct function of what I put in. The threshold of what I can achieve is raised slightly, the speed with which I can achieve it is raised quite a lot - but it's not a magic rocket ship that whisks me to far off hostile environments and protects me while I'm there. I'm still "naked". The code is still mine, and I can get off the bike at any point and walk.
dTal··on The Meta Glasses backlash is changing how (or if) people use them
1) You can't use a GoPro like everyone else?

2) You can't use Bluetooth earbuds like everyone else?

3) I'm sure your infant child will appreciate their representation in Meta's training data.

dTal··on Don't you mean extinct?
They predict the "most likely" token given the context. That's a huge caveat. Just putting "this is excellent code" in the context makes a vanilla LLM do better. Doesn't that make you pause for a moment before asserting hard limits on what they're capable of?

You might argue they're still capped at the "best" quality seen in the input. Not so. Take typos. Human text has a certain base rate of typographical errors. LLM output contains almost none. Why? Because there are many more ways to be wrong than right. LLMs are not just averaging machines, they also denoise. That should also give you pause.

dTal··on Abject Praise
You appear to be arguing that we shouldn't care that Apple provides a consistently bad experience and locks their users into it, because web developers can level the field by degrading everyone else's experience too. There's a nice couple paragraphs in the article that explain why we should in fact care, and you don't seem to have addressed them, so let me reproduce them for you in case you missed them. (also fyi "it's not that hard, ask your LLM" comes across a bit snarky):

These large, persistent gaps matter to the mobile and web ecosystems because Apple is unique in denying access to more capable, less-buggy engines and actively erecting unlawful barriers to choice when forced by legislation to enable it. This is accomplished through eye-watering budgets for legal shenanigans, direct lobbying, and well-heeled astroturf front groups to maintain a capability gap between web and native.4

That chasm is instrumental in trapping users and developers in the extractive vice of Cupertino's App Store. A persistent, material gap in capabilities creates a perception of the web being less-than; a budget option for the unserious. Should users choose more capable, more private, less buggy browsers for a larger share of their computing needs, Apple might lose the leverage that enables it to extract rents.

dTal··on OpenBSD has a use-after-free allowing local privilege escalation to root
>you can't really criticise an AI lab for doing AI instead of straight up giving money to security researchers

Sure I can, if they - or you - pretend "the entire point" is about useful security work rather than expensive loss-leading marketing and demand creation.

We shouldn't look at this and think "wow AI is super useful for security". We should look and this and think "wow, there's a LOT of capital going into persuading us that AI is super useful for security".

THAT is "the entire point".

dTal··on Capitalism Gone Wrong
> I am sure Facebook did not begin as a project designed to steal people’s time. It began as a cool app that people genuinely loved.

  Zuckerberg: Yeah so if you ever need info about anyone at Harvard
  Zuckerberg: Just ask
  Zuckerberg: I have over 4,000 emails, pictures, addresses, SNS
  [Redacted Friend's Name]: What? How'd you manage that one?
  Zuckerberg: People just submitted it.
  Zuckerberg: I don't know why.
  Zuckerberg: They "trust me"
  Zuckerberg: Dumb fucks
dTal··on Capitalism Gone Wrong
I'm curious how he rationalizes the thriving market for nicotine patches.
dTal··on Capitalism Gone Wrong
1. No. It was always awful for an unacceptable number of people. It's just your turn to feel the pinch. I suspect you'd get a different answer if you asked someone from China.

2. We have to look forwards, not backwards. Yesterday contained the seeds of today. Today contains the seeds of tomorrow. No point rewinding unless you want history to repeat.

dTal··on EU Parliament greenlights Chat Control 1.0
Conversely, declaring a space "politics-free" tramples a hell of a lot more than curiosity.

Incidentally, the comment immediately below this one begins with the words "I'm curious". So I think we're good.

← PreviousPage 4 of 34Next →