ai-skeptics 2026-10-14

Let's first solve basic real-world problems :-)

👍🏻 1

Cal Newton is good at explaining LLMs and breaking through the hype. He's not a Luddite, of course. 😉

Great way to start a day 😅 Posthuman Hitler wannabes in Highlander standoff solving the universe. “There can be only one!” - my universe. Resulting in Star Trek future with no humans or bio in it. Sounds great Just a normal day this days 🙂

@geodrome I wouldn’t call it crazy. I’ve had similar thinking when I was 6, and it did sound great in my head and talking with my 6 year old friends. We had such a great time! In kindergarden… If somebody invested in us that would have been so great. My point is that it isn’t crazy, it is dangerous

Well it can be both crazy and dangerous 🙂

💯 2

Could providers pick winners and losers among companies and individuals by serving different models to different customers under the same name?

That would be an interesting way of doing market manipulation

Right, and very hard to detect, I would guess. It would show up as anecdotes mostly. A large swath of people would have to surrender account access to researchers for there to be any type of reliable statistics on it.

They can even do it according to statistical distributions. Pretty impossible to detect if you don't have all the data

I'd be surprised if it wasn't already happening at that one lab with a messiah complex.

They admitted this week that Houthis used Claude to help their missile guidance software. Anthropic's internal safety team is almost certainly exploring ways to nerf models for specific customers in hard-to-detect ways.

Not to mention the existing work they’ve done to curb distillation, but from what I gather, that’s a more in-your-face “no”.

I mean I thought the plan was once everyone is a meat-proxy you can control the markets by changing what the meat proxies proxy.

You also get an oracle for free

If people do baseline what LLMs tell them

then you can ask the LLM to know what people will do in a given context.

Kinda ties into all Darios work on predicting how neurons fire.

In some ways if your competitors use LLMs you can know what their market strategy will be (by asking the LLM the same questions they would ask the LLM)

Seems like AI is a tool after all, just not in the way we expected.

Not jut for company , but mass manipulating people or countries also become options for indirect control

Fachidiots

Yikes! They are putting the "cult" in silicon valley culture

I've heard rumors of coworkers in the valley spending down their 401k because they wouldn't need it with AGI around the corner, this is a step more 😕