redox99

Correct me if I'm wrong, but it didn't actually generate CP, it changed people into bikinis, not actual nudity.

You can go to the beach or (I assume) search on Google and find people of any age wearing bikinis. And that's fine because to normal people (not pedos) it's not something sexual.

I understand that someone specifically editing a minor into less clothing is probably a pedo. Probably the same way someone generating feet gets aroused with that stuff. But that doesn't mean a model generating feet is problematic.

I do think making the model reject the bikini thing is a good idea once you find out it's happening, but let's not pretend it's the same as the model generating actual sexual content.

It doesn't.

Obviously the missing part (which we already have for software and math) is that we need agents to be able to run automated loops in the real world. That basically requires robots. I think we'll be there in less than 5 years.

Usually at least here it's tens of thousands of laws like "replace article 3 paragraph 2 with:" and it's not like you can ctrl f "lemonade stand" and get something useful.

A strong llm without that knowledge can probably spend millions of tokens parsing every single ordinance and give you an answer if you directly ask about that.

But both a local human or an "all knowing" llm when you just tell it about the stand will be like "hmm wasn't there a law about low calorie beverages being tax exempt? Let me check" and after a very precise search come back with a tip that if you sell sugar free lemonade that's tax exempt.

I mean would you get an accountant from another country just because he can use search?

The human genome is 800MB, (and 500MB with compression according to GPT) so that part is easy.

I think the problem is that the conversation drifted from "all useful human knowledge" to "enough knowledge to do my job".

Both a human and a current large model will bring up unprompted some tax regulation that applies in your district that could matter to your task. A small model won't know the tax code of every city in the world, as it's probably impossible to fit, and it's the kind of thing that unless you already know about, it's very hard to search for, let alone come up with unless specifically prompted about something tax related.

Unless we start having different small models trained for a certain profession for a certain city, I think we'll need those trillion parameters.

I think a specialized model could squeeze all you need to know about a certain profession in 30GB. But not all professions at once, which is what these models try to do.

Or maybe not, maybe there's a world model needed for human level at any profession that is very hard to quantify and requires more than 30GB by itself.

It's definitely trained on generating SVGs. Not that they trained on the pelican prompt itself, but probably a more general RL step that has it generate SVG and a stronger vision model judges it.

It's up to you if that counts as benchmaxxed.

(mumbles) Shannon entropy... Kolmogorov complexity... something, something...

On a more serious note, it depends on your cutoff for "entire human knowledge". It's easy to prove for a generous interpretations of "entire human knowledge" that it can't be done, but hard for something like "all useful human knowledge".

Yeah but results are worse, see https://news.ycombinator.com/item?id=49301574

Asking it factual information[1]. You just can't compress the entire human knowledge into a 30GB file.

[1] Without searching the internet. And even if you allow it, you'll get much worse results because search means browsing and parsing the top results, and search results are horrible, whereas internal knowledge from training encompasses the entire internet plus all books including very niche stuff.

Yep. These small models are actually worse than GPT 3.5 at some tasks (like recalling facts). You can definitely make models smarter at specific tasks (like tool calling, coding) but you can't compress the entire human knowledge into a 30GB file. It's just not enough bits.

Benchmarks mean very little. The difference between Luna and Sol in the real world is massive.

Almost surely the free model on the web ui. So a dumb model, no harness, and probably a bad prompt.

The cache reads are really insanely expensive.

Yeah [1] is really a thing and SlopCodeBench (https://www.scbench.ai/) kind of measures that.

You need to manually push models to clean up the slop every now and then otherwise it becomes chaotic. And every change with LLMs is always extra lines.

I'm sure Grok 4.6 is not Fable level. Benchmarks are almost useless.

Having said that, Grok 4.6 (1.5T params) is without a doubt way smaller than Fable, maybe a Fable sized Grok would be Fable level?

Yeah the speed was very nice indeed.

That's pretty surprising. Idk about Grok 4.6, but Grok 4.5 was clearly below Fable, Opus 5 and GPT 5.6 Sol.

It's a bit worse.

I haven't tried so it's pure speculation based on benchmarks, but I'd assume Grok 4.6 is around Opus 4.8 in real world use, but clearly below Opus 5.

They weren't sold out, they had bad seats or didn't have 4 contiguous ones.

For a Linux user, you can already build such a system yourself quite trivially by...

Computer use will suck until we get 500+ tk/s

The internet sucks. Yesterday, I had my agent search for openings for The Odyssey that fit my requirements and then book them. It was way better than manually looking at seat maps for 20 different showtimes and going through 10 steps just to buy the tickets.

They're improving but you definitely need both a strong AGENTS.md and also usually many LLM passes (one for implementing a feature, another one for code quality, large passes every now and then for major refactors, etc).

That's not really a typescript thing though, just an LLM thing.

Obviously sleep is necessary. I'm saying that there isn't always a large pressure to minimize sleep. Long sleep can be beneficial just to preserve energy.

That one for some reason is [flagged]

You definitely see a bit of this in other primates. They sleep on trees (much safer) and for a lot more hours than humans.

Do not, my friends, become addicted to sleep. It will take hold of you, and you will resent its absence!

I think you overestimate the disadvantages of sleep. It's very species dependent, for some it might be a disadvantage because you're more vulnerable to predators, for others it's an advantage because you spend less energy. If you're safe you really only need enough wake time to eat and reproduce.

It wouldn't matter if it was stored only on their servers. As long as they offer the feature to downgrade a chat to a dumber model that can be jailbroken (and the downgrade keeps the reasoning), this trick works.