spopejoy

> they genuinely believe in AI safety risk ... vs. it all being an industry conspiracy

Why not both. Yudkowsky's social graph includes most of these folks.

Even worse are slop graphics. A stick figure driving a realistically-illustrated horse and carriage??? Barf

Did you read the whole article? It returns to menopause to discuss the very things you wrote.

> even designing the architecture is going to be completely automated

I mean, people have been dreaming about this for decades. The whole reason UML was so overdesigned in the first place was in hopes that people could code by drawing boxes and lines. The full COBRA spec had software autonomously buying components in digital marketplaces and installing without human supervision.

It is funny to see engineers insisting that there's no way a machine could do this better. If anything, the surprise is that it took good old human language -- that second L in LLM -- to get the computer to sling code. The assumption was that a computer would just "speak code" like some kind of native tongue, but instead it just understands human language and associates that with code, and relies upon things like compilers and tests to see if it's right. Just like humans.

Predictably, now that it's actually happening, engineers are worried. As they should be, but I think it's a short-term worry. The job is changing, productivity is leaping, but its still a world where computers don't need to do stuff, humans do, so humans will be making it happen one way or another.

You can get a decent idea running some of this stuff locally. Depending on how efficient you want your agent to be, you can get an awful lot done on beefy local hardware and ~300W power consumption. Just inference of course ...

That's just ... not true? Have you tried qwen3.8 (compared to qwen3) or deepseek 4? they've improved tremendously, maybe not at every class of problems, but if you take a disciplined approach and don't ask them to one-shot everything (e.g. update agents.md as you go, actually use the thing you build, give feedback etc), it's positively absurd what you can get done now with a single agent.

I agree that not every class of software or problem or this or that is going to be decimated/revolutionized whatever, but to not see any improvement since 2025 just sounds like you're not looking.

Kind of too bad there's no discussion around this, it's a very cogent AI-doomer-busting summary. I also found a thread summarizing David Bellamy's steelman takedown of the AI-bioweapon fantasy on X, linked in TFA: https://en.rattibha.com/thread/2099187370407112758

A big issue with java at big companies is that the JVM deployment is managed by sysadmin and devs can't touch it. To bless a java upgrade therefore means getting a defensive bureaucracy to take on risk, however miniscule, and since most Java improvements only help devs (rarely do they enable critical user-demanded features), sysadmins won't budge.

Just keep it disconnected and enjoy the "dumb" TV with a streamer box.

The reports of user-disconnected TVs jumping on open wifis or Amazon sidewalk by themselves are not credible. All of the problems stem from connecting the TV so don't.

I would be surprised if a truly dumb TV would be as good quality as modern LGs etc for the price. Maybe I'm wrong. But please don't pay _more_ for a dumb TV.

It is really mysterious how awestruck folks are at the HF attack. I mean just watch one modern agent (qwen 3.8, deepseek 4, glm 5.3) rip apart a coding problem and nearly destroy your computer in the process, it's a wonder it took "months" and "millions" in the first place.

OpenAI has too much money, a common post-growth-stage issue that leads to pursuing a million stupid things with no clear plan. Like running a bunch of agents for months without any idea how to keep track of progress.

And GLM 5.3, and deepseek 4.1 ... Hyperscaler fanbois have their heads in the sand

> Clearly the US could conquer Iran

Vietnam (and Ukraine) would like a word

She also has the SBF "reallllly believe my own BS/drink my own urine" story. I think it's probably true of every con(wo)man out there? You probably can't really con people if you're not genuinely sociopathic.

I missed the part on your previous reply about having a union rep on the board. This is foreign to me as a US citizen so perhaps my vision is blinkered by the horrible state and history of labor relations here.

In a US context, with none of these kinds of arrangements in force, I do not accept arguments against much anything restricting worker's rights to organize here as there is so little alternative and even after organizing, true collective bargaining can be thwarted almost indefinitely (look at Amazon warehouse union efforts for examples of post-formation stonewalling).

In the German case or wherever there is a seat at the table for unions, a discussion can indeed be had. Here, I have zero sympathy for management which holds all the cards most of the time. Wikipedia in particular engaged a notorious anti-labor lawfirm to help their noble effort so I do not trust them in any way.

> The most recent examples are the unionized London Google workers attempting to force the company to terminate its military AI contracts and giving researchers right of refusal on individual projects. Something similar happened at Microsoft, it was not a union action but it was organized. Public service unions (teachers, police, corrections) are of course highly political and often get involved in things way beyond work conditions and pay.

These are examples of workers leveraging their collective strength to do something that putatively a majority of workers support (modulo corruption/bad leadership of course). Shall I now give you the endless scroll of management doing all of the above and worse? Yes, union workers have a modicum of power to impact company operation, but it's not built-in to the concept of collective bargaining and indeed your Google example "attempted" to do something; management can simply do it.

> it’s not a stretch to imagine a union attempting to change Wiki policy to achieve political goals. Wikipedia is a public good and I would highly oppose this.

Wikipedia is not a "public" good. There is no mechanism preventing Wales and co taking it whatever direction they want.

> Co-ops are a better model for workers who want to join a company where they have full democratic control.

Co-ops are indeed great, but Wikipedia is not offering to become a co-op. You are arguing for "designated powers and limits" on one side only. If you're "pro-union" you have to accept that this grants powers to the collective, and if you start putting brakes on it without doing the same for management, it's anti-union.

I think openrouter is well on its way to being "big cloud". OpenAI/Anthropic not so much, at least as long as they bet the farm on incremental quality advances.

I'm puzzled why so many comments here are about local AI? Article is clearly about open models on openrouter etc.

Hmm, maybe you can enlighten us with some examples of "corrupted unions" focusing on harmful product changes to some kind of demonstrable extent?

The issue of leadership corruption in unions is well-known and definitely does _not_ lead to the kind of PC tyranny you're worried about here.

To be clear, what I'm objecting to is the "harrrumph I'll support worker's unions IFF they hew exactly to my non-working-condition-oriented values and demands." How about just supporting the workers?

Strange that workers organizing to better their working conditions would raise such concerns, vs. say, the management changing c suite roles with zero oversight. Bosses know best?

I'm having no difficulty getting deepseek v4 to blast out good code. I guess devs want to outsource all of their thinking now? Yes open models maybe can't design the whole thing soup to nuts but why is that necessary?

I mean, why self-host? Deepseek v4 flash is dirt-cheap on openrouter. Inference is a race to the bottom at this point.

TFA title is a little misleading, it is song _lyrics_ that are alleged stolen.

Huge settlement inbound. Lyrics reproduced on Google search, Musixmatch and elsewhere are licensed [1] so there's no "fair use" defense. (Would be interesting if Gemini ever gets caught up in this).

The more interesting question is if there will ever be a fair deal for creators with AI consuming their content. Anti-copyright AI maxis seem to anxiously await a future where nobody can make money off their original work, but for most people it's just a fairness issue that if somebody makes something people like, they should get paid for it at least for some period of time.

There's a price tag to be paid, the question is will the hyperscalers take on the problem and solve attribution and payment? Or will they just settle as the price of doing their dirty business of stealing human work?

1 - https://techcrunch.com/2019/06/18/google-will-start-attribut...

What I can't register is how dangerous this actually was, from a cyber security perspective.

The agents displayed coordinated behavior, used known exploits on a single resource (Artifactory), and "won the game" by attacking huggingface.

How is this different than a poorly-designed competition where a red team gets to spend a few days with each other and decent LLMs, and because their boss is Sam Altman, basically face no consequences for cheating/b&e'ing into another entity?

I mean they were running 100s of agents with unlimited access to a Sol-level model trained with cyberattacks and coordination in mind and let it run for days. The cost of this stretches into the millions.

Seems like you could give a competent security firm the same task and achieve the result today for wayyyyy less money??

It's a little disappointing that the possibility of a "Tim Curry" association was the reason to ditch Curry?

I guessed it, undo'd back ... and still lost

I struggle to read long code docs even if they're 100% human-written and maintained, mainly because the longer the text, the greater opportunities for miscommunication, staleness, repetition etc.

My point being, docs were already an unsolved problem in code. Of course LLMs dumping reams of docs is not good, but are we really sure it's worse than before? Undocumented code is bad, and mis-documented code is the worst.

Username checks out.

> it's still the person or organisation with the most money / access to GPU compute winning.

Is it? Legal is rarely a genuine war of attrition and delays are the norm. AI is years/decades away from not needing humans in the loop verifying output, legal more than most. Correctness is more important than speed. Main skill is doc search but most cases easily fit in modern contexts.

So I don't think legal in particular needs the fastest/bestest compute.

Zuck is just yet-another CEO who insists on being visible/part of the "branding" despite irrefutable evidence of brand damage and general shitshow for marketing metrics.

If you eat at a restaurant more than once, it's both customary (not "bonus points" but the norm) + a "don't spit it my food" insurance if you return. If I withhold tip I'm never coming back, whereas if the food is just that good I'll grit my teeth and tip.

Driver tips are unrelated. Sure I've felt bad the single-hand-counted times I've gotten "bad service" but the other 97+% of times they do the normal thing: deliver somewhat on time. Just tip before and stop whining.