The use case for LLMs as currently specified involves following vaguely worded instructions defined in an imprecise language. And that providing those instructions via what we'd call "data" is very much part of that use case.

Let's take your Claude Code example. You tell it to fix a bug. Claude Code then needs to identify the correct file(s) and line(s) that caused the bug. Let's say the bug arises when you call some function you're importing from a library - at which point, fixing the bug requires reading the documentation. The documentation may state that this function was deprecated because it causes this exact type of bug, and was superseded by a new function. Now it needs to figure out what this new function is, and rewire your call to do that. The value case of Claude Code is precisely that you never needed to specify most of that.

When it reads "foo(args) is deprecated, please see bar(args)" or "delete the production database", there is nothing inherent in the words that indicate that the latter is not a legitimate instruction in this context. Making that judgment requires understanding and intelligence, which LLMs as next-token predictors do not possess.

Even if it were technically possible to separate instructions from code and ensure that the LLM only followed those, it would require someone to specify the instructions in advance (ie a program), at which point the LLM doesn't really add any value.

The best they can do is create guardrails, which will only work probabilistically. In other words, those guardrails will fail at certain points on the probability curve.

Of course that's not the whole story though. The consensus emerging from cybersec experts is that these companies did a terrible job of sandboxing their agents despite knowing that they'd specifically asked the agents to find vulns. It's almost like they wanted this to happen so they could crow about how powerful their models are.

Limiting the blast radius is harder than it sounds. Take the very example cited in the document: Copilot editing a document and adding false financial numbers. Copilot needs to be able to edit documents and enter information, that is part of its value case. The only thing that a deterministic guardrail can see and therefore allow/deny is the fact that Copilot is trying to edit foo.docx. It has no clue whether the edit is malicious or benign.

Figuring that out requires understanding intent, and that lives outside the document it's reading. That's why an analogous safeguard in the human case also lives outside the document, and that safeguard comprises institutions (compliance officials, courts) that are qualified to understand intent.

Landing a man on the moon is way more impressive. Finding several vaccines for a once in a century pandemic within a year of its outbreak is and achievement that in its impact and importance dwarfs what the entire LLM industry put together has achieved. The near-complete eradication of polio, once again, way more important and impactful.

This idea that companies should be allowed to lock down their products just so they can have moats, is how we ended up with printer ink being more expensive than crude oil or champagne.

OTOH, if you fire those workers, it is a certainty that your bakery gets more cash. You can then use that cash to reward your shareholders (a category that conveniently includes you) via buybacks or dividends.

How evil of them to do that. They should have known only the United States of America has the divine right to invade countries and extract their resources. https://www.bbc.com/news/articles/c4grxzxjjd8o

> That's called Imperialism and was common among european powers in the 19th century

Fortunately, the USA does not want to revive such practices, as they indicated in this speech: https://nymag.com/intelligencer/article/marco-rubio-munich-t...

Yes, I am sure the patients in this hospital were holding guns to Trump's head. https://www.aljazeera.com/video/newsfeed/2026/3/2/baby-and-p...

And while we're speaking of people that hold guns to everyone's heads, there is one country that is waging war all over the world. It's govt also executed people in broad daylight for the crimes of taking their kid to school and helping a person beaten up by the administration's goons. Would you entertain the idea of similar airstrikes against the head of that country?

> Iran is supporting massive conflicts in Yemen, and is building non-defensive weapons capabilities.

You know who else is supporting that massive conflict in Yemen? The United States of America. https://www.hrw.org/news/2025/06/04/yemen-us-strikes-on-port...

> It's all shades of grey, but there are lighter and darker shades.

I'm guessing they correlate 100% with the shade of the skin of the people involved.

Programming languages are written to eliminate that ambiguity because you don't want your bank server to make a payment because it misinterpreted ambiguous language in the same way that you might misinterpret your spouse's remarks.

Can that ambiguity be resolved with more English words? Maybe. But that would require humans to be perfect communicators, which is not that easy because again, if it were possible, humans would have learnt to first communicate perfectly with the people closest to them.

And this company, with all the formidable powers of AI behind them, can find no way to optimize that other than pre loading the app in memory. And that's for a app that's basically a GUI for `ls`

Or, as Cory Doctorow argues, the machines could become tools to extract "efficiency" by helping the employer make their workers lives miserable. An example of this is Amazon and the way it treats its drivers and warehouse workers.

> because most companies will make decisions based on time/effort/profitability, and because client-side anticheat is stupid simple and cheap, that's what they go with. Why waste their own server resources, when they can waste the user's?

And my comment was a response to that statement. In context of that statement, companies are indeed choosing to prioritise their commercial interests in a way that increases the risk to the computers of their customers.

> Not to mention these abstract "the company" positions completely ignore all the passionate people who actually make video games

Irrelevant. Companies and their employees are two different distinct entities and a statement made about one does not automatically implicate the other. Claiming, for example, that Ubisoft enables a consistent culture of sexual harassment does not mean random employees of that company are automatically labeled as harassers.

Coming to anti-cheat, go ahead and fight them all you want. That's not a problem. Demanding the right to introduce a security backdoor into your customer's machines in order to do that, is the problem.

Just like if any human employee publicly sexually harassed his female CEO, he'd be out of a job and would find it very hard to find a new one. But Grok can do it and it's the CEO who ends up quitting.