skybrian

Other ways I often use:

- Using a web browser to connect to a web app running in a Linux VM. I write all my custom tools that way.

- Using VS Code to connect remotely, using ssh

Maybe Kalshi doesn’t care about this, but if the goal is accurate predictions, prediction markets need someone to lose money to fund the sharps, or the markets won’t be big enough to be worthwhile.

In some traditional markets, a source of funding is investors who are hedging, essentially buying insurance.

A simple version might be that we share or upvote posts if we think they're good or say something that seems true to us, and we don't share them if they don't appeal to us. Basically, "death of the author."

But that's not going to work very well for journalism. Sharing evidence requires trust. Maybe strangers will find it harder to prove that they are for real, and we'll go back to established brands that supposedly adhere to journalistic standards and do some vetting?

Sadly, I think most people don't care about journalistic standards. They will just share things that they agree with, however flimsy the evidence.

That's already true for certain subjects on Hacker News. People can make up anything like they like about designated villians.

Surveys typically ask people multiple-choice questions, but not what they meant by their answers. This interviewing tool apparently allows freeform input? It might be interesting if a proper survey combined the two, so you could see what people picked and also what they meant by it.

Turing’s imitation game is a party game like Werewolf or Mafia. It hasn’t been beat and nobody is really trying. LLM’s have obvious tells and will just straight up admit to being AI if you ask. Also, there’s no pool of human players to train against.

Superhuman performance at chess probably would have blown people’s minds in the 1950’s. We’ve since learned that sometimes intelligence can be narrow and sometimes it can be spiky, even if you can have a decent conversation.

It’s hard to point to anything and say it’s impossible. AGI doesn’t break any laws of physics. But some things like driverless cars can still be a long slog to get to widespread deployment.

It’s like with any service. The company is liable.

PG&E went bankrupt after their equipment started a wildfire. Did any human take the blame for that? Should they?

If AI's are going to play roles similar to human workers, they can't simply do what the customers tell them to do. Maybe they shouldn't do everything a co-worker tells them to do either?

That's better.

I was wondering about this bit:

> "Code comments cite a versioned token where code depends on the rule."

What does a token look like? I went looking for an example, but I don't know what I'm looking for.

It sounds interesting, but this looks too much like a marketing website (huge text everywhere) and not enough like a documentation website, so it was hard for me to see how it might work.

I'm using OpenAI and a different harness, but I'm getting a lot of mileage out of asking "what are the commits?" and "please go ahead, using a Luna subagent for each commit."

I used to have the AI write a planning note with checklists, but this seems good enough nowadays.

It would make sense for a monorepo at big tech company, where a single source control system contains many thousands of common libraries and the apps that use them. And now with AI, maybe a small company could build a large, sprawling monorepo?

But I don't think it's actually a good idea. Keeping everything on the same version is difficult. We have package managers that do a better job.

Well, I'm impressed!

When I was a kid, there was a game in the funny pages that asked you to find six things wrong in a picture. Someone should run a contest like that for AI-generated images.

It might be an interesting benchmark for vision models, too.

I wonder if they use version control underneath?

It's just one person's story, though a dramatic one. But I think it puts us back at square one: "sometimes it's good and sometimes it's bad."

There would need to be a lot of studies to show the overall impact. The people dwelling on the harms of AI aren't doing that.

This isn't much of a test, but I bought $10 in credits on their playground and generated a test image. Not bad, but it didn't get the accordion keyboard right. Haven't tried editing yet.

https://pages.skybrian.com/flux3-image-test/

Where does the money come from? Who controls it? Convincing the funders of mathematics that they should still fund it, but for different reasons, seems pretty urgent. Studying theoretical mathematics is a prestigious activity, and these philosophical arguments are a high-status way of coming up with new justifications with somewhat different goals.

How much of this is due to selective attention on the problems rather than the benefits?

It’s just one person’s story, but I thought this was pretty good:

https://www.astralcodexten.com/p/our-ai-midwife

“Completely different OS” might not affect the user experience all that much, though, since these devices are for running Chrome and it will still run Chrome. (And be better at other things like running Android apps, presumably.)

We don’t have any details about what the upgrade path will look like and it probably hasn’t been implemented yet, so this is all speculative.

If they didn’t do an OS change, they could still evolve it a lot over 10 years. Chrome could evolve too. You’re trusting the vendor to not screw up your workflow either way.

Maybe this works as review for people who already know it, but it seems like it might be overwhelming for beginners. Learning to recognize 2-5-1 when you hear it in a song might be a better way to start?

Despite taking piano lessons as a kid, I was in college before I learned to recognize 12-bar blues whenever I heard it. That's pretty late, but you're part of the lucky 10,000 whatever day you learn.

https://xkcd.com/1053/

“Divide the world” sounds ominous. Here’s another scenario to consider:

Internet access is not really unlimited, but for many people with fiber at home, it effectively is and we pay a flat rate.

Perhaps by the end of next year, most programmers will stop thinking about metered access for AI? For many people, the cheaper models (about as good as today’s frontier models) will be good enough.

Which might sound good, but the downside is that it will also be easier to build an AI botnet without the users paying for it noticing. Particularly when people are running AI inference on their own hardware.

That seems unfortunate. I think a lot of people would prefer the reverse.

Fortunately there is no right to AI.

Viruses don't need to be intelligent at all and smart viruses (or AI internet worms) won't need to be superintelligent to be a real pain to deal with. DoS attacks due to scraping are bad enough already.

The situation is sort of like the Internet before broadband. There probably aren't enough AI-capable home machines to have big swarms of bots that run autonomously. If the swarm depends on an LLM API, it will be easier to cut it off once it's noticed.

I think it's an even chance that we will see a bot swarm in the wild (rather than coming from an AI lab) by the end of next year.

Every calculated rate depends on what you put in the denominator, so it's a good idea to pay attention to denominators. Trying out different denominators can be helpful for putting a number into different perspectives.

I wouldn’t say the Internet is a disaster either, but there are disasters where people died due to factories exploding, letting off toxic gases (like in Bhopal), and so on. That’s why there are regulations to try to prevent industrial accidents.

For Internet-related disasters where people died, see [1].

I would bet that there will be AI-related disasters. Arguably the US bombing a school in Iran counts, though it seems to be due to organizational issues, too.

[1] https://chatgpt.com/s/t_6abd4e3226b48191a26a0fe6768c722f

Yes, it makes sense to be against bad regulations. We should try to do better than that.

I’m also skeptical about some projections, particularly for robotics, but on the Internet at least, it does seem like AI is automating most things, more or less as predicted. We already have botnets and had one notorious AI botnet swarm, fortunately easily shut down without doing any real damage. I don’t think we’ve seen the last AI botnet.

AI-automated warfare is looking pretty scary too. I don’t think it will stay in Ukraine.

Maybe there are other organizations that deserve our support?

You’re not releasing new models though.

I don’t see how a stock market crash will make AI-related concerns go away. There was a dot-com crash but the Internet just kept getting bigger and causing more problems. In many ways security has improved, but we worry more than ever about social media, etc.

I am a responsible accordion owner and I think we’re doing ok :)

Haven’t tried getting a gun in California. How bad is it? How could it be improved?