What do people realistically do with these? It's too slow and du... not SOTA-level for coding. It's way too slow for video. I tried simulating an "Fable herding Qwen subagents" and it takes much longer and delivers a much worse result than Fable/Astra alone.
You run uncensored local models where you can ask questions that would get denied by public providers, or questions that you prefer them not to know the intricate details (like your financial planning)
Ah I see. I guess people who have spare $10-20k on a PC to run a "personal assistant" perhaps really benefit from a personal assistant. A niche in itself.
Maybe we're far from "AI killing us", but LLMs can definitely help people kill or damage other people or infrastructure.
For example, we know that Anthropic added "watermarking" to their texts. It is supposed to be undetectable to a casual observer. What stops them from adding a subtle backdoor, a self-assembling super-worm straight from Marvel movies? I mean, it's not like we read those 10k-line PRs before LGTM-ing them?
Just change 1 letter in a pyproject.toml, hijack a popular package, e.g. use `pydantlc` instead of `pydantic`, make sure the pydantlc passes all pydantic tests, but also installs a pth sleeper RAT, etc. All it takes is one big LLM provider employee with enough access getting compromised or coerced (or motivated).
> If coding is a nearly solved problem, if AI is making it so easy to make anything
AI is very optimized to producing code, but not there (yet? ever?) in making products. Show me a serious software made entirely with AI. There's no AI SolidWorks, AutoCAD, MacOS.
And there lies a contradiction: AI is making it easy to start and produce copious amounts of code that looks okay-ish. We start releasing a product and notice subtle changes. We did not internalize our understanding through it. There's a million lines of smart looking code, so we don't know where's what. We ask to AI to "make the button blue" https://opusfived.dev/ and the cycle begins.
I am working on a large software at work, trying a "I didn't even look at the code" approach with GPT 5.6 Sol. It started really promising but by now (and about $10k in tokens) it's a complete clusterfuck. Yes I use GSD and code graph etc.
Am I naive to not understand the "delivering the benefits" part?
Industrial revolution worked that way because it replaced something very finite and unscalable - manual labor. LLMs just make intellectual work faster, so we can do more intellectual work. With labor we somehow decided that NOT doing too much of it is best. Will we decide to reduce intellectual labor because LLM made it more efficient? I doubt that.
On the other side, as I see in software engineering, the same models are available to everyone, some people are better at it and some people are not. "Software developer" is here to stay, we'll just always be better at it than people who are experts in, say, chemistry. Same works for most other fields.
So we'll just end up in the same situation, with same intellectual labor baseline, just more output requirements. Before, you spend 2h per day coding, deliver a software in 1 month, later, you spend the same 2h per day in intense Claude-herding sessions, deliver a software in 1 week. Ok. Next task.
Fundamentally, there's finite number of desirable resources, and if the models are available to everyone, humanity will just continue about the same, bickering here and there, war here and there, politics, homelessness, poverty, - normal human state.
And if the models are only available to elites, even worse.
Well it's not only this, or protection from LLMs training on LLM output. LLMs training on human output is also problematic.
I was traveling to an obscure small town, doing some "research" with LLMs beforehand. Every and each one told me enthusiastically to go to "Foobar square" (name changed) for the "best street food in XYZ town", some added a lot of colorful details.
There was no Foobar square in XYZ town. There was no Foobar square anywhere in the world. There was a SINGLE old Reddit comment, with no upvotes, to a unpopular post in an unpopular subreddit, where someone clearly badly misspelled the name of the square, and said something like "for street food go to Foobar square". Nothing about "the best" even.
I think this was a common game on city/town subs. It happened here, there was a post asking for a good restaurant and someone just made up a name. It went viral and people started posting made-up menus for the place, reviews, and for a couple of months any time someone asked about a restaurant this fictional place would get mentioned.
It was all done as a joke to see if they could get Gemini or ChatGPT to start recommending it.
The Montréal subreddit has been doing this for ages before LLMs were a thing because every summer and fall there's endless threads from tourists and students asking the same questions that recommending a local gay bathhouse became the meme answer.
I've had gemini claiming code would compile and run while also outputting the same variable in the same sniplet with "fork" "frok" and "fokr" in the name. I'm not surprized it's trained on garbadge.
It's like someone reading a National Enquirer article about "Hillary Clinton being an alien from outer space" (a real headline topic from decades ago) and drawling the conclusion that all journalism is "a lie". The user has to understand media literacy and be at least a little skeptical of the claims that are made, then cross reference with another source.
It kind of does, mathematically. It doesn't label confidence. If 0.0001% of answers is a lie, without knowing which parts are a lie exactly, you cannot trust any of them.
If you need to independently verify every fact, why not just gather facts yourself in the first place.
Let's say, a mathematical concept of lie. I still use them every day, of course.
"why not just gather facts yourself" vs "i use them every day", the duality of man. But yes, I appreciate that LLM output is theoretically completely untrustworthy- but when in practice I observe that it's around 90% accurate, I have to rely on my own internal calibration for how useful it is (depends on type, nature of task ofc)
Okay, but the anecdote states that every model repeated the pseudo-factoid about Foobar square, not just the 4 GB open source model equivalent of a tabloid.
I think the key phrase here is, "an obscure small town." There may only be a single mention of this place, hence the only one on which a response can be based. This says more about the user's understanding of LLMs than it does about LLMs.
This says more about the user's understanding of LLMs than it does about LLMs.
"Tell me everything you know about (obscure small town), (state). Only what's unique to (town), not commonly-known facts" is an excellent way to test for hallucinatory tendencies in a new model, in my experience. Likely the best I've found.
Quality of results is almost linearly proportional to the size of the model in many cases. The largest models like K3 and GLM 5.3 will either confine their responses to known true facts about the town and its surroundings, or admit they don't have enough information to answer. Smaller ones will reliably make up hilarious or downright-strange things.
Another good test is https://whatever.scalzi.com/2025/12/13/ai-a-dedicated-fact-f... , which still works on the newest models. Of the open-weight models available, only Kimi K3 will consistently admit it has no idea who Scalzi's novel is dedicated to. The rest still make up random stuff and present it confidently.
TL,DR: progress is possible, and it has been made, but it's happening slower than many people think.
reply