dTal

The point of the comment you are replying to is that it's also not worth the time and effort to use LLMs to find vulnerabilities, if "time and effort" can be measured with "money". If you factor in all the money spent on training, GPU data centers etc, it's not actually a financially efficient way to find bugs unless you profit from creating demand for LLMs. LLMs aren't cheaper than humans per unit work, yet. They're just massively deficit funded because capital thinks "AI" is going to reshape the world order, and wants in.

It does more types of charts, including Gantt charts. Very useful.

if you have a 2 dollar budget to make a viewfinder work, just throw in a ten cent plastic lens and be done with it.

>LLMs use a neural network to interpolate

"interpolate" in what vector space, pray tell? What does "interpolate" even mean, when I prompt it "write me a story about a sentient banana in the style of Hemingway and oh make it a commentary on class consciousness"? You can't assemble such a thing by cutting and pasting pieces of other text. That kind of "interpolation" has to happen at the semantic level - ipso facto, there is a semantic level.

Not to mention that no, they don't predict the next token in the training set. Give any LLM the first paragraph of any Wikipedia article - almost certainly in the training set, and uniquely so - and it won't predict the next word correctly, a lot of the time. But it will predict a word that is grammatically correct, stylistically apropos, and most likely factually correct. So what's it really doing, hm?

LLMs aren't even large enough to contain their training data - not even remotely close. It can't "stitch together things it saw" because it doesn't remember them. It only remembers the ideas used to construct them. The learned abstraction is the entire point of the exercise. LLMs would be useless if they were overfit the way you say they are.

Rats are a weird choice for that comparison. They're some of the smartest animals on the planet, some say smarter than dogs.

Are you sure? It's popular to call them "ineffective" because it's a safe way of saying they annoy you while virtue signalling that you do in fact care about climate change and do think that something should be done.

But I've never heard anybody follow up that complaint with their own, more effective suggestion. In the general case, disruptive and particularly self-sacrificing protest does work. It's how women got the vote.

They do not "stitch together" anything. Neither on a technical level, nor a philosophical one. It "scales better than you expected" because your mental model is wrong.

And, not to insult you, but it's quite obviously wrong. As a mental model it fails to explain basic capabilities. How can an LLM follow elaborate instructions? How can it respond appropriately to user input, when the user input doesn't match any previously seen text? Hell - how does it even balance parentheses? There is no way to explain any of this without conceding that the LLM has semantic understanding. It knows that this comes after that, but "this" and "that" can be at an arbitrary level of abstraction.

Sure - they generate text "like" text they've seen before. That "like" does a ton of heavy lifting.

I was idly browsing F-Droid yesterday and found a Claude coded B-REP CAD program built on OpenCASCADE, all touch friendly and everything. Definitely something happening.

No thanks.

>We don't need kernel level anti cheats.

More than that, they'd be illegal by your own standard, if applied fairly. You can't steal control of someone's computer just because they bought a game from you. That's a strictly more severe "exploit" with strictly more impact on "quality of service" than sending bogus data to a server. Cuts both ways.

But we all know how it'd really play out. If lying to a server counts as "hacking", but corporate rootkits and locked bootloaders are fine, all you've really done is solidify "felony infringement of business model". You haven't protected anyone. Video games are a stupid thing to immolate computing freedom for.

This already exists more or less exactly as you describe. Family mobile phone plans offer the ability to designate a number as having parental restrictions, and the phone company will intercept DNS requests and substitute a version of Google with SafeSearch enabled and locked, provided by Google for this purpose along with a friendly message "your network operator has locked your safesearch settings etc etc". Don't let anyone tell you this isn't mainstream.

Mobile phones rather famously tend not to be plugged into landlines.

So quite a lot then?

They're resigning because Google is complicit in killing people. How is that "identity politics"? Do you just throw the word "identity" in there for no reason?

Uh huh. And why is it crazy?

>If you think nuclear facilities are complicate I would suggest to look into facilities used to manufacture polycrystalline silicon for solar panels, or facilities for manufacturing gas turbines used in gas power plants.

That's an unfair comparison. We are talking about operational costs, not capital costs. A fair comparison would include nuclear's capital costs, which don't do it any favors - nuclear plants also need fancy turbine blades.

Not that I'm against nuclear, I think we're completely mad to be still burning (burning! so primitive!) ancient plants, coming up on a century after Magic Energy Rocks were discovered. "Cost" is a fickle metric when so many costs are externalized in both space and time - nobody cares if the effects of pollution are felt years down the line, thousands of miles away.

But it's funny that governments throw wildly generous subsidies and special legal treatment at domestic food production, correctly perceiving cheap reliable food as upstream of a functioning society, yet fail to similarly "overinvest" in energy sources for machines as well as humans. I hazard that most of the European countries that depended heavily on Russian gas would not have been so blase if their population had been subsisting off Russian food imports.

We should be building out tremendous energy capacity using every conceivable technology available. Any country that does this is virtually guaranteed wealth, for energy is fungible with almost everything else. The equation is so obvious that failure to do so implies regulatory capture by entrenched energy interests. "Too cheap to meter" belongs in the same rhetorical bucket as "Perfect sound forever" - a promise that was retracted as soon as it became clear that scarcity was more profitable than abundance.

If your language is indeed the first to be born in a mobile phone, after all this time, that would rather suggest that there is a difference. Ergonomic differences and artificial restrictions are still differences and restrictions.

that you know of

Bits is bits - so sorry you couldn't figure out the right bits to make the hardware do what you want, but those bits do exist and they are findable in principle - it's still all software! Your lack of knowledge of the signing keys is qualitatively no different than your lack of knowledge of any other undocumented proprietary information required to program the chip. Frankly you'd struggle even with the keys.

Infeasible? Oh my yes. Impossible? No. That's the point - there is a distinction.

Bad take. Some things are feasible and some things are not, "anything is possible" is a useless framework. Example: go convert two smartphones to communicate p2p over their 4g radios - it's all software!

LLM "awareness" is similarly irrelevant. They process information usefully, in a way grounded in reality, and that's that.

I dunno, Chez Scheme is from 1985 and remains today one of the most magically optimizing compilers for a dynamic language in existence... kind of makes you wonder how we went so far wrong with Python.

ollama is just a wrapper around llama.cpp, and a pretty janky one at that. You're much better off using it directly.

Because the local LLMs available today are already fantastic, and the difference between no LLM and an open weights LLM is much smaller than the gap between an open LLM and a so-called "frontier" model.

It's important that people get used to the idea that your interactions with a language model are a highly personal thing. LLMs can perceive and categorize us in ways we can't even imagine, far more violently than the simple algorithmic feeds which have already corroded public discourse so much. LLMs can control us. LLMs warp the information landscape more radically than even the internet did. Even now you are likely underestimating their role in future society.

The principles of software freedom are becoming existentially important.

Link? You have to tell us if you want to break through the noise!

We aren't? Are you sure? Where is your email inbox? Where are your backups? Where are your music files? For most people the answer to all those is "someone else's computer".

Anyone with an M-series Apple computer can run something very competently. Mac Pro users can run 30B class models which is good enough for the vast majority of practical everyday purposes, far better than the original ChatGPT was. Anyone with a gaming computer is in a similar situation. The rest of us can still run stuff, just not as big or as fast.

It does not, but it might encourage more people to care. Worrying about training is a luxury when you are starting from a baseline of "OpenAI spies upon me and controls my access". Let's focus on getting every Tom, Dick and Harry 1) on board with LLMs, because they're happening, 2) habitually using local software.

It's good that there is a movement for open LLMs, but it's not where the battleground is right now. The battleground is local vs service LLMs, and we are losing that battle badly despite all the software being here now and viable, entirely because UX sucks.

How many normal people do you know who use "ChatGPT"? A lot, probably.

How many even know what "Gemma" is, let alone have downloaded llama.cpp, a GGUF file from Hugginface, and run "llama-server" from a text console with all the correct command arguments? How many are thinking about this use case when speccing out their next computer? Where is the breathless marketing copy boasting x tok/s?

We are sleepwalking into slavery.

Ah, I think it's safe to say you wouldn't. Nothing against you, but the personality required to acquire a trillion (!!!) dollars is incompatible with the kind of philanthropic thinking you clearly possess.

Take this as a sign that you're out of the loop and your bubble is smaller than you thought. niri is the current hot tiling compositor, and I've switched my laptop over to it recently - away from Plasma - because it's so incredibly slick. It's gathered enough mindshare to have several desktop environments explicitly target it (Dank Material Shell, Noctalia).

In fact there's a whole new cultural wave within open source - tiling Wayland compositors, shockingly featureful DEs based on QuickShell, brand new TUI tools with no CLI or GUI equivalents, and most of it written in Rust.

I think this is a kind of nerd chauvanism. What I see is that the general public are deeply skeptical of "AI" in all its forms. Software "engineers" are especially vulnerable to believing that LLMs are smart generally, because LLMs are good at writing code, and skill at writing software is how the software engineer measures the superiority of their own intelligence. But a poet is in no danger of over-anthromorphising an LLM.