crthpl

They were using it in May

what specifically has changed about Claude Code?

it does not use haiku. you can just ask Claude (possibly not fable BC of cyber) to reverse engineer the obfuscated JS.

For the US models, you can look at the rewritten CoTs or just ask them what they think is happening.

where the heck are you seeing OpenAI racing to IPO before anthropic?

on the API, their margins are very high

no, it is discounted expected future cash flows

Yes it can! That's the whole point of RL! it generates slightly out of distribution rollouts, and rewards good rollouts to change the distribution of the output

where the heck did you get those parameter numbers from?

are you sure it's 60 fps and not uncapped?

have you tried pangram? it's basically the only good AI detector, and they have nearly 0 false positives

Anthropic is very big (the biggest AI co?) in B2B, where you don't have ads. Also, if they end up creating a datacenter full of geniuses, ads won't make sense either.

1. you can definitely tell apart an S-curve and an exponential if you look at the derivative(s). AI progress does not seem to be close to the middle of the S-curve. 2. e.g.: a slowdown hasn't shown up in moore's law yet.

It is the case that Anthropic employees have no usage limits. Some people do experiments where they spawn up hundreds of Claude instances just to see if any of them succeed.

the chain of thought is what it is thinking

The reason they get a perfect score on AIME is because every question on AIME had lots of thought put into it, and it was made sure that everything was a possible. SWE-bench, and many other AI benchmarks, have lots of eval noise, where there is no clear right answer, and getting higher than a certain percentage means you are benchmaxxing.

GPT-4o is so far behind the frontier; you shouldn't use it as an indicator of what LLMs are capable of.

Another reason may be that WebGPU didn't allow for as much optimization and control as Vulkan, and the performance isn't as good as Vulkan. WebGPU also doesn't have all the extensions that Vulkan has.

If humans were too complicated to be mathematically modelled, then we wouldn't exist.

Why would charging for extended support hurt Microsoft? They "win" if you pay for it, or if you upgrade to windows 11. If you do neither, you'll probably get hacked down the line if you don't airgap, and Microsoft doesn't gain or lose anything (except goodwill, but that doesn't matter with the kind of monopoly they have).

I think there is a deterministic algorithm (based on GPS position?) for determining who climbs and who descends.

I'm guessing MacOS, Linux, Android, Windows, iOS, and Web?

The idiom "Don't judge a book by its cover" applies to many situations, but books are not one of them.

dawn is the WebGPU backend in chromium, while wgpu is the WebGPU backend for firefox written in Rust. wgpu is seeing a lot of use in non-browser uses; there are some examples on their website.

https://wgpu.rs/

The 7 million is probably how many different hues we can see. We can see many more different brightness levels.

Not if only one species came to earth.

Or perhaps the other way around: we are social because we have language.

Their Performance section shows that the 12GB is 50%-150% faster than the 3080 Ti (so better than 3090 Ti!), and it also has more cuda cores and memory than the 3070.

Eye candy is one of the main reasons people build Lego bulids.