Oh OK. Murdering rich people might effect social change. Though I’m skeptical because not a lot of people actually do it, which is an important component of that change.

I think you still end up coming around in a circle: in your fantasy of mob-justice, what prevents the wealthy from ending up being the ones who control it? (Like Vlad the Impaler!)

“Exactly as horrible” is unfair to you. But it sounds like you’re advocating to give up on rule of law and just have the strongest most violent people be the ones to decide what’s right. And I’d argue that you’d just get back to where we are right now: wealthy people would control the system, they’d employ strong violent people to enforce their personal whims as “law”, and you’d be complaining that nobody is willing to beat up the pedophile (because his friends would hire goons to kill them).

I mean, presumably that’s what’s stopping you personally from implementing your own recommendation, right? Because if you showed up and kicked this guy’s ass you’d be beaten and arrested by the police.

That I’ve unwisely installed some bad software and I should delete it, as recommended.

Do you feel like you understand what “hacked” even means? If so: a phone is a little computer. If not… I donno. I guess you should give a bit more story about why you’re interested in “hacking” and yet know so little about it? Something that makes it more interesting to reply.

For basic questions, why wouldn’t you start with Wikipedia, which is an excellent resource? https://en.wikipedia.org/wiki/Phone_hacking

It’s so resilient, too. Like, I can see not examining things that Zuckerberg says because they don’t sound idiotic on first glance. And the US is very much “more money = better than” so why not give extra weight to anything some rich guy says.

But Musk and Trump say multiple obviously-stupid things every day, are visibly wrong all the time, and still get breathless treatment. “Wow, Elon Musk says he expects oil to go to $69/barrel by the end of the week! Does he know something others don’t??”

Maybe Katana is somewhere with stronger consumer-protection laws than the USA. It’s not possible to make judgements about legal/illegal without at least knowing the jurisdiction.

We can say for sure that it’s obnoxious and unethical.

Your example sounds fine to me, but I wonder if maybe the real value of AI coding that you’re seeing is not “AI writes code” but rather “AI ignores all the coding rules/conventions, which were being badly applied in this case.”

Like, I could picture trying to write a simple tool as you describe and running into the wall of “We’re an Angular shop so you need to use the framework.” But the AI isn’t forced to follow these rules, so it can do a decent job.

This has been the main value of AI to me. It’s just like, “sorry, the machine won’t let me do a refund,” as an alternative to listening a customer scream that they deserve a refund.

Even this is disappointing. LLM bullshit is only impressively fluent compared to older generative systems. (It is very impressive compared to them. It just should have stayed in academia longer and its components could develop into useful things. Instead everyone’s falling over themselves about a kick-ass demo.)

Now what do you want me to do? Ask an non-deterministic LLM to implement the code from scratch every time I need it in my project?

I have some coworkers who are excited about exactly this and I don’t get it at all.

OK but I would hi-five those people. It’s harder to fight capitalism if you’re also fighting health problems!

Right?? It’s bizarre to me that otherwise-smart-seeming people will think they can write “explain your reasoning” to the AI and it will explain its reasoning.

Yes, it will write some fluent response that reads as an explanation of its reasoning. But you may not even be talking to the same model that wrote the original text when you get the “explanation”.

I asked plenty of questions on SO and never had a bad experience. But I put quite a bit of work in. You couldn’t ask “how do i sort a list in JAVA” and get answers, you had to ask “here’s some code I’m writing <working example> and it does <x> but I think it should do <y> because <z> what’s going on?” and people gave some really nice answers. (Or you could put “how do sort list java” into web search and get a fine answer to that; it’s not like SO was the only place to ask low-effort questions.)

One of the bad things with AI is it’s soooo helpful that when I get questions now it’s like “please create a DNS entry for foo.bar.baz” and they’re asking because the AI got completely stuck on something simple (like making a request to api.github.com) and wandered up and down and eventually decided on some nonsense course of action and the developer has given up on thinking about anything.

They did not edit past commits (that causes tons of problems). You can easily find the earliest commit that credits Anthropic. Example https://github.com/lutris/lutris/commit/6d5f18eb92e558ffda37781386812f1e9f1a2499

Have you tried using AI

“We live in a capitalist hellscape so why bother to struggle against it” isn’t a completely unreasonable take. Too bad, though.

I donno anything about China, but whoever made this meme certainly doesn’t know anything about the USA. The idea that “liberals” or anyone else (??) are high-fiving themselves over a credit score. lol

As a developer: yes to the developer and data scientist and data engineer. Scientists and engineers should be responsible for their work.

The BI analyst: maybe, if they’re responsible for collecting data that ignores the impact of the service on teens. If they’re doing sales-comparisons between Anthropic and OpenAI… eh, I donno.

The janitor: probably not since I don’t feel like the deaths are widely publicized and they probably work for a contracting company that handles the building.

The “agents” and “agentic” stuff works by wrapping the core innovation (the LLM) in layers of simple code and other LLMs. Let’s try to imaging building a system that can handle a request like “find where I can buy a video card today. Make a table of the sites, the available cards, their prices, and how they compare on a benchmark.” We could solve this if we had some code like

search_prompt = llm(f"make a list of google web search terms that will help answer this user's question. present the result in a json list with one item per search. <request>{user_prompt}</request>")
results_index = []
for s in json.parse(search_prompt):
  results_index.extend(google_search(s))
results = [fetch_url(url) for url in results_index]
summarized_results = [llm(f"summarize this webpage, fetching info on card prices and benchmark comparisons <page>{r}</page>") for r in results]

return llm(f"answer the user's original prompt using the following context: <context>{summarized_results}</context> <request>{user_prompt}</request>")

It’s pretty simple code, and LLMs can write that, so we can even have our LLM write the code that will tell the system what to do! (I’ve omitted all the work to try to make things sane in terms of sandboxing and dealing with output from the various internal LLMs).

The important thing we’ve done here is instead of one LLM that gets too much context and stops working well, we’re making a bunch of discrete LLM calls where each one has a limited context. That’s the innovation of all the “agent” stuff. There’s an old Computer Science truism that any problem can be solved by adding another layer of indirection and this is yet another instance of that.

Trying to define a “limit” for this is not something I have a good grasp on. I guess I’d say that the limit here is the same: max tokens in the context. It’s just that we can use sub-tasks to help manage context, because everything that happens inside a sub-task doesn’t impact the calling context. To trivialize things: imagine that the max context is 1 paragraph. We could try to summarize my post by summarizing each paragraph into one sentence and then summarizing the paragraph made out of those sentences. It won’t be as good as if we could stick everything into the context, but it will be much better than if we tried to stick the whole post into a window that was too small and truncated it.

Some tasks will work impressively well with this framework: web pages tend to be a TON of tokens but maybe we’re looking for very limited info in that stack, so spawning a sub-LLM to find the needle and bring it back is extremely effective. OTOH tasks that actually need a ton of context (maybe writing a book/movie/play) will perform poorly because the sub-agent for chapter 1 may describe a loaded gun but not include it in its output summary for the next agent. (But maybe there are more ways of slicing up the task that would allow this to work.)

As with dating apps, if the mental health recommendations actually worked then engagement would go down. I doubt they even need a human putting their thumb on the scales; the algorithm optimizes engagement in whatever way it needs to.

It’s basically impossible to tell with these between the example being totally fabricated, true but only happens some small percentage of time, true and happens most of the time but you got lucky, and true and reliable but now the company has patched this specific case because it blew up online.