I can imagine some government prompting an LLM to break into some other governments system to steal state secrets.

”Here are the files you were after. Btw, I also stumbled across the nuclear launch codes, which I’ve confirmed are working. Was I a good machine? :3”

deleted by creator

Such law need to clearly define what development towards ASI actually entails. An easy workaround is to just play along with the narrative that LLM is a hopeless dead end and can’t possibly bring us closer to ASI.

I believe the cases of ”math problems settled by AI” will slow down. Anthropic probably want some high profile math problem at the same level as Navier-Stokes, but after that I don’t think there will be as much progress.

They got the publicity stunt they wanted. What more is there to gain? Navier-Stokes costed many millions to settle, and probably was close to being resolved by human mathematicians either way. Who else has anything to gain by spending ridiculous amounts of compute on true/false answers?

My guess is that the AI companies will focus on finding publicity stunts in other fields. Possibly medicine.

I bet this AI agent won’t last 1 year until they rebrand or discontinue completely.

One is thinking about the sky. The other is thinking about the ground.

I know! Let’s divert energy and resources to torture. That will maximize the benefit of humanity.

• Some brilliant CEO mind

Here’s a tip to make yourself popular at a party: walk up to someone, take a picture, and ask the AI to roast the person. Tell it to be nasty and use forbidden words. It’s going to be so fun!

His job is probably easiest to replace with AI. Just make up random bullshit and tell investors what they want to hear.

Probably depends on the country, but it’s easily solvable with meat proxy until constitution changes.

I think at least one country in the world will have elected at least one political party openly run by AI within the next 10-20 years.

People will grow increasingly angry over politicians inaction regarding people being replaced by AI, so they will vote for the party that will replace politicians with AI as revenge.

Probably be some European country.

US will settle with an AI Trump as a facade for the Republican Party just to keep MAGA going.

It’s probably going to get worse.

Soon there will be an influx of randos on social media who just vibe through every open problem in the book with no regard for research ethics. If OpenAI have no regard for research ethics, why should randos do it?

But hey, at least it will be formalized in Lean alongside with some gibberish written in LaTeX.

If you find a controversy page on some Christian pastor, then you know you’ll need eye bleach after.

Societal collapse it is then!

If that ever happens I see three possible options:

• Universal Basic Income
• Communism (money is abolished)
• Societal collapse

Just did that. I’m a professional at this point.

Mixed feelings. I’m kind of an AI pilled doomer.

I see that AI can be a productivity boost - if used correctly in the right places. But I’m also increasingly worried that it might one day replace any work that’s done on a computer (which is most white collar work).

I’m doing a lot of data processing work. I’ve been able to see AI do all this in one go:

• Read a vague Jira task item about some data error.
• Diagnose on real data what the underlying problem might be.
• Find the line of code that need to be changed.
• Rerun the data pipeline with the new line of code.
• Verify on new data that the error has been fixed.
• Submit a pull request describing the change with a detailed report on impact analysis.

Is it perfect? No, it’s often making mistakes and digs deep into the wrong rabbit holes. But is it crazy that it has managed to do all of this once? Yes it is. This was unimaginable just a few years ago.

If this level of automation going to continue in this pace, what’s left for us humans?

When the Christian friend convinces you into religion, but it turns out God hates belief in false gods more than non-belief (you picked the wrong religion).

A clanker will use any permission you’ve granted it.

In this case it’s a two line change: change to round down instead of round up. I suppose it’s the latter where the AI assisted him in identifying which line should be changed.

https://github.com/torvalds/linux/commit/818bebeb63dd6bf5f4e07e145f6cdbace520a34c