sho_hn

I've commissioned art before via the Krita web forum's job section. Lots of bright young artists, many of them students, that do great work for affordable prices to build their portfolios.

You don't really need to be in the 1% to be a patron if you need some art.

This one is much better: http://www.youtube.com/watch?v=tKrVYOUvrtA

> The cost of raising a child is enormous.

Honestly, the primary problem is still the mindset behind most times this line is written.

The cost of raising a child is enormous only if you consider the return a nuisance at best, which many seem to do. If you consider it very rewarding thing you want for your life, working for that a little doesn't seem all that preposterous anymore.

The same peope typically have no problem spending money on frivolous hobbies, toys, travel etc. they technically get no economic return for either.

There may be people who are economically barred from affording children, but most people set the bar at "I want kids with no strings attached".

The real reasons are that we've created a society in which many people no longer consider a rewarding experience they actually want, and those who don't want it no longer have it externally thrust upon them by cultural expectations or pressures as much. I think the latter is good, but the former may be a pervasive failure of character.

> Shocked I would ever see this again on HN of all places

Wouldn't HN be one of the most likely places?

Link/Cloud gets you some more things than just the Remote Access part.

Nabu Casa also hosts e.g. STT/TTS services for a HA to use. Again this is easy to replace with something else and entirely optional to begin with, but since I trust Nabu Casa more than, say, Google and don't have the hardware for local models I like to use them.

I think the "supporting HA development" argument is really not a bad one. HA is so much better than anything else by such a wide margin.

What "migration"? This is just a rename.

Once you have payment processing in-house, the decision of what devices to support is yours to make.

If you don't, two lock-ins and network effects reinforce each other.

I read Lane's The Vital Question this year and enjoyed it quite a bit. If you've read both, would you compare them for me?

I've tried here and there, but even the top-end models always tend toward shallow generalist takes. I've not been able to get more than basic primers out of LLMs, and nothing close to the ability of a professional author to stay on course with an idea, detail level, specificity etc.

People in the early days used to often whine that LLMs just regurgitate text snippets (unfounded of course), but I think the way we currently train and RLHF them actually seems to largely make them unable to reproduce the knowledge they have been trained on, since they seem to just always want to please the mean with their output. I'm oversimplifying the mechanisms, but you get my drift.

I honestly think it's more likely that younger generations grow up seeing the ill effects of being AI-pilled for communication/writing, deciding by way of AI literacy that they want none of that, and develop better ideas on responsible use.

I really don't like AI for document writing as a whole, especially within organizations. Code can be a form of communication, but for the most part it's an engineered thing that can be re-run, tested, verified and so on. But documents drive decision-making, and decision making is often one-time. Decisions should be studied and understood, and depriving authors of much of both is not having any good effects.

I'm not an academic, and I understand by osmosis that organized science has many problems related to publishing and connected mechanisms, but I have to say there's been quite a few moments in my life and career when I found the inspiration or solution I needed in a good paper.

For condensed knowledge on highly specific topics (i.e. more specific than is economical in book form) I often know nothing better.

Most recently I did sort of a self-taught crash course in ground-penetrating radar and applications for an archeological endeavour I write software for, and a number of papers have been really invaluable.

"Zig with TypeScript bindings and React components" for a TUI is definitely the wildest, weirdest stack for a given job I've seen.

Dunno, honestly, most HN people have an extremely narrow view of what constitutes "tech" (= web commerce and adware) and wouldn't survive 5 minutes in an actual engineering company. Europe is all over the supply chain of just about everything that is hard to make.

It's not.

What a strangle "Europe is failing!" comment thread, when ASML is a European company selling to all of the factories.

Yep, that's what I meant!

Definitely!

I'm giving a talk this Sunday at KDE's Akademy conference of our use of KDE software in research at the Great Pyramid, if anyone's curious. It's a short talk slot and I cannot go very deeply into detail, to keep expectations low, but you can see cool photos of robots.

It's vibecoded. This is Codex' favorite layout for some reason.

I'm (obviously, I think, given my initial reply?) fully aware of this, and I think it's entirely besides the point. "They" don't need to have a goal to emergently cause a problem, and the inability to "focus" over long periods can be moot when you have swarms of runs exchange and mutate state, as in the HF attack.

Intent or how intelligent LLMs are doesn't actually matter. Even if you just treat it as a sort of fuzzing attack that can be biased/weighted better than other fuzzers, or bumbles around with a statistically greater likelihood to "strike cybersec gold" than other algorithms, we've never before seen organizations run things with such a large potential outcome space with anywhere near this kind of compute before.

I think it's actually kind of the dismissals that are usually overly emotional or biased toward treating "LLMs" differently. If in some kind of alternate universe simpler genetic algorithms would have had these properties and we threw similar amounts of compute at them we could have the same conversation.

The things the OP listed mostly aren't particularly wild. I think it's you making them out larger than they are, and therefore more unlikely, which is why I take issue with your original comment.

> running on the hardware they started on

They just need to acquire a payment method and rent some infra, and exfiltrate their own data. Or pay another provider that hosts the same models already. API calls.

> being able to be turned off

You can reasonably equate this to "saving state across executions", which the message board attacks already did.

> having limited computing power

Renting more infra, variant of the above. API calls.

> "not repurposing resources currently in use for other things" (like the atoms in your body)

Ok, the "atoms in your body" bit is a bit silly, but making API calls to put physical resources into play (even if it's just, say, ordering something on Amazon to somewhere) is of course easily possible.

None of these is in complexity much different than the HF attack.

This grossly understimates the risk, imho. The problem with LLM runs is that people run programs without knowing the outcome beforehand, with a large potential set of outcomes unlike any other class of program we've run at this scale before. In the interaction with other systems (since we also give them far-ranging access, very nice hardware, and run them often), bad things can happen.

It's like running potentially buggy code - or an well-biased fuzzer -, but at massive scale, and code that can self-modify and self-expand. "Alignment" is just a way to describe aggregate statistics about their runtime behavior.

They don't need to be intelligent, or alive, or "more than token prediction engines" for this. They just need to happen to end up making the wrong API calls without the operator seeing it coming. No virus has a brain, yet they can be very bad for you.

I understand that some people get turned off by anthropomorpization or scifi language. Fine! But don't turn off your engineering brain over it.

I usually throw out the tiny thorn.

Like my parent said, the grapes in East Asia are thick-skinned, and people either peel them or suck out their flesh. The ones sold in Europe are very thin-skinned, and peeling is largely impractical. I'm relating the latter to boba.

> from what I can tell.

The authors are not RubyGems. The website says it's based on data served up by RubyGems. They point at OpenAI with arguments.

Did you try very hard "telling"?

Considering RubyGems was part of the HF story, seems likely to be connected.

Indeed.

Although what keeps me up at night is the worry that it's easier to automate attack than it is to automate defense, and that containing these systems is a losing game. Could an optimally competent OpenAI succeed?

I occasionally commision visual art from young artists/students via Krita's web forum, and what I find to be true is that if I had AI do the same art, it would just be exactly what I ask the AI to do or iterate toward (I'm detail-oriented enough, and iteration rounds are cheaper and more plenty). If I ask a human instead, I get the somewhat-unpredictable input of another creative mind besides my own brain, and often this sums up to something greater.

It really depends on what you want/need, augment/extrapolate your own creativity with AI tools, or invite someone else's.

The AI debate is for now most often centered on whether AI is capable of replacing individuals, but replacing the results you get when you put multiple humans together is another dimension entirely.

Ouch, yes. One time I commissioned a musician I had befriended to make a piano arrangement of a favorite song to be played at our wedding. The result was great, but it sure wasn't cheap, even if reasonable.

I'd still do this again however, because I really enjoyed the feeling of the community of people contributing our wedding and all the little stories attached. "I prompted ElevenLabs for this!" wouldn't be a memory; a fellow human spending some hours of their life on making this for us and what lead to it is.

There's often value in the provenance of things.