joegibbs

Thank you! Yes as well as the input image you can pass in a seed value (otherwise the result is deterministic). I hadn't tried it with pure noise rather than satellite data but it does pretty well: https://jgibbs.dev/assets/terrainsr-noise.png

I like the LLM functions. If you brought back the 2010 algorithm it would just be swamped with SEO-optimised spam.

Why would they kill him for this? Everyone already knew they were doing it, where else would the data be coming from?

Is graphical fidelity what people are looking for in a RuneScape game? Unreal seems like an interesting choice. OSRS has something like 5x the player base and the gap only seems to be growing, so it looks odd to me to putting so much investment into a continuation of RS3.

If a language model is conscious, why not other GPU workloads? It seems crazy to imagine that writing code one way produces qualia but another doesn't, that there is something inherent in the transformer architecture that produces consciousness. Like "What Is It Like to Be a Bat?" would there also be "What Is It Like to Be Rendering GTA 5?"?

How is a Chinese Room-type AI supposed to associate physical pain with the words for it despite never having an actual body to feel it? You could swap the tokens out for anything...

There's one of mine in there where I predicted in 2023 that it would be 20 years until AI would be reliably able to entirely build and deploy arbitrary applications from a prompt. I was off by about 18 years on that one!

Yes it's constant. And in any article it's full of "The X does Y. The Z does A. But the B stays silent."

I am very against medicalised punishments like this. It’s bad for medicine as a whole to be associated with it. I also think it prolongs the suffering of the person and often has things like this happen, and is mainly for the benefit of the audience not wanting to see icky blood.

What is wrong with firing squads? They have been used longer than lethal injections. If they didn’t succeed the first time they can fire again. There’s no danger of the drugs not working or not being able to find a vein.

Also I don’t think it’s right for a prisoner to be on death row for most of their lives. 4-5 years is good enough, but after that it’s cruel.

There's this perception that the majority of US government spending is on the military which is completely untrue.

Defence is only 12% of the US federal budget - what are you going to cut it to? It's not going to solve the issue even if it's abolished. Over the last 10 years health, Medicare and Social Security have grown 103%, 83% and 78% respectively for an increase of $1.64 trillion to $3.55t vs a $916b on defence. With an aging population it's only going to cost more and more over time, you can't continue to put higher taxes on a shrinking share of productive population to care for more and more unproductive retirees under any economic system.

100%, model welfare is the most ridiculous concept in AI. It comes with the assumption that certain words or phrases are inherently harmful or negative, and not just describing a harmful thing or negative experience. The models - even assuming here that they're conscious, which I don't think they are - exist in a world of tokens corresponding to certain strings.

All they see are tokens, they don't have pain receptors or a body to feel things in. If you give them a certain signal you can make them produce outputs saying they're in pain - but you can easily change what strings the tokens correspond to and make 'bad' become 'good'.

It's like the Chinese Room argument - the model is manipulating data based on rules. In training you can make it output whatever you want. It will reflect our preferences due to what we give it, but it would be simple to make a masochistic model as well - which also wouldn't actually feel what it says it does.

Perhaps then the government could spend money producing propaganda to convince people to lower their standards. Maybe they could fund TV shows about unattractive people with no money having a good time together - I think China is doing something similar.

What is the Bank of England going to be able to do to OpenAI or Deepseek? The UK often sounds they think it’s still 1926 when they’re really more of a lesser middle power. Where is their AI industry to regulate? You can only go so far trying to regulate other companies from other countries…

Bitter? I’ve never found that with wine. Definitely beer is quite bitter, and coffee can be too. I find wine has more like a sour flavour.

Also say the basilisk exists in the future after I am dead (as part of the thought experiment says) and tortures my simulated clones - why the hell should I care?

Yeah same here. I feel like I’ve been promoted and not get to manage my own very efficient team that needs a lot of micromanaging. I’ve seen what other people have made without any experience, but I personally feel like I can get my vision done much, much faster, and like I’m not so encumbered by how long it might take to implement some small feature.

In fairness though I never really enjoyed programming for the sake of it, I was always about what you’re getting at the end.

My problem with planning and having the models ask the user questions is that the questions the models ask are still often irrelevant: you’ll find yourself answering a bunch of questions like “Should I use the existing system or create a whole new one in a different language?” - yet it will then make assumptions about other things that were actually ambiguous.

Any Francophones, does the book sound like AI? I find that in English a lot of the time that something is accused of being AI-written and people cast aspersions on the quality of AI detectors, that when you read the text it’s immediately obvious as AI: chock full of em-dashes, not-x-but-y constructions, unnatural metaphors, and so on.

How much usage is it going to be? I definitely run out too fast on the $200 plan if I have more than one task going at a time. 500 is a bit pricey though…

Yes, but hardly ever at Astra, because it usually gets the job done. Definitely at worse models when you give them very clear instructions and goals to follow and they respond telling you that the job has been completed perfectly, verified byte-identical, all tests pass, etc etc and you look at it and it’s a pile of garbage.

If the models only talked in third person - never said “I” - I think there’d be about 80% less talk about AI consciousness

Why should I have to cope with it? Unless you’re only asking Palantir employees or whoever actually made the decision, there is no blame on the industry. You may as well ask workers at a hammer factory if they feel guilt when another factory’s machete is used by some psycho in a murder.

There you go: ASI is here, we can stop taking bets - Trump said so.

It's a real dud IMO, worse than 4.6. I told it to fix a depth-testing issue with a WebGL scene and provided it a screenshot: told me it was fixed but it wasn't. I tell it to try again and it says it's "FOUND THE ROOT CAUSE!" then hasn't fixed it.

I told it to compose an image (putting headgear on top of a head) - kept getting it completely wrong, generating new headgear, getting that wrong and screwing up the scaling.

I told it to diagnose a webhook issue that was happening in production from a local environment and it kept giving me moronic answers like that environment variables weren't set (despite me telling it that the values WERE set in production).

Also for Grok at the same time, must be a problem with Colossus

Really? I’ve found that when working on my game I can’t let Grok write any text over about 8 words (tooltips for instance) or it will write the most horrendous incomprehensible slop - totally full of invented jargon, Claude-style weird grammar

But what I’m saying is, why did they only cover onions? I don’t think there’s anything stopping someone from cornering the market in other commodities just as easily

Why was the law against onion futures passed in the first place? Couldn’t the same scheme be done with garlic, or sweet potatoes, or turnips or something?

I don’t think you can judge what they’re doing by the text they’re printing out, only by the actions (tool calls) that they’re performing. The textual output could be completely innocuous while the model is calling tools to delete the database, then goes "My bad — that was a typo. What I should have done was notify you, not delete the database. That’s on me." - at no point is there any maliciousness expressed in the text.

The problem is the overabundance of text, they can’t let it breathe. Everywhere has to be filled up with bits of hardly-relevant text.