nl

39 ms·

661.
by nl 8mo ago
I don't see how you get that reading. Anthropic is clearly allowed to sell Claude to companies not doing business with the US Military. If anything that's more likely to be non-US companies.
662.
by nl 8mo ago
I think a lot of the conflict about what imperialist policies means is different framing. For better or worse, inside this the border in this map China has fairly imperialist policies. Outside it not so much: https://en.wikipedia
663.
by nl 8mo ago
> This is true of packaging and build systems in general. They are often the passion projects of one or a handful of people in an organization This is a very insightful comment
664.
by nl 8mo ago
RAM is pretty different to Huawei 5G base stations. Australia for example is a large and growing market for Chinese electric cars. China is the biggest export market for Australian raw materials so it doesn't just put random trade barr
665.
by nl 8mo ago
DUV machines are ok, but it still takes 2 years to build a clean room factory.
666.
by nl 8mo ago
> The only way they will is if the hyperscalers and AI companies start to implode -- which will kill a huge portion of the US economy and lead to global recession, so, cheap RAM but nobody can afford it This isn't right. RAM prices
667.
by nl 8mo ago
Sure, but the practical form of this attack is limited. You can't maliciously embed it in a site you control to either steal map usage or run up their bill because other people's web browsers will send the correct host header. Tha
668.
by nl 8mo ago
Does anyone really believe something like this? There's a long stretch from over optimizing a UI to something that is very clearly an error like what has happened here.
669.
by nl 8mo ago
Llama 3.1 8B is pretty useful for some thing. I use it to generate SQL pretty reliably for example. They are doing an updated model in a month or so anyway, then a frontier level one "by summer".
670.
The Pentagon's UFO Psyop(garberfiles.substack.com)
3 points by nl 8mo ago|1 comments
671.
by nl 8mo ago
Taalas is interesting. 16,000 TPS for Llama on a chip. https://taalas.com/
672.
by nl 8mo ago
I had a very odd interaction somewhat similar to how weak transformer models get into a loop: https://gist.github.com/nlothian/cf9725e6ebc99219f480e0b72b3... What causes this?
673.
Stress testing Claude's language skills(vivsha.ws)
2 points by nl 8mo ago|0 comments
674.
by nl 8mo ago
This is pretty interesting!
675.
by nl 8mo ago
Simon's work is always appreciated. He thinks through things well, and his writing is excellent. Just because something is popular doesn't make it bad.
676.
by nl 8mo ago
This doesn't seem anything to do with AI (except in the broad sense the account getting disabled seems automated)
677.
by nl 8mo ago
> Our second model, still based on Taalas’ first-generation silicon platform (HC1), will be a mid-sized reasoning LLM. It is expected in our labs this spring and will be integrated into our inference service shortly thereafter. > Foll
678.
by nl 8mo ago
Releasing this on the same day as Taalas's 16,000 token-per-second acceleration for the roughly comparable Llama 8B model must hurt! I wonder how far down they can scale a diffusion LM? I've been playing with in-browser models, an
679.
Chat with Llamma 8B at 16,000 TPS(chatjimmy.ai)
3 points by nl 8mo ago|0 comments
680.
Llamma 3.1 8B in hardware, 16,000 TPS(taalas.com)
4 points by nl 8mo ago|0 comments
681.
by nl 8mo ago
Why not both? Scaling laws are real! But they don't preclude faster processing.
682.
by nl 8mo ago
There's a reason Google model usage on OpenRouter is so high - it's easier to pay the OpenRouter tax than it is to figure out how to pay Gemini directly.
683.
by nl 8mo ago
Codex has been post-trained for coding, including agentic coding tasks . It's certainly not impossible that the better long-horizon agentic performance in Codex overcomes any deficiencies in outright banking knowledge that Codex 5.2 h
684.
by nl 8mo ago
It's more like saying AWS has a monopoly on virtual machine hosting. (For those unaware, AWS doesn't have a VM monopoly, and the market dynamics seem similar)
685.
by nl 8mo ago
I'm not sure that's what they meant. Read like this: > the tractor and plow are doing a lot more work than 3 farmers, yet we've got more jobs and grow more food. it sounds to me like they mean "more job and grow more
686.
by nl 8mo ago
> Last I checked, the tractor and plow are doing a lot more work than 3 farmers, yet we've got more jobs and grow more food. Not sure when you checked. In the US more food is grown for sure. For example just since 2007 it has grown
687.
by nl 8mo ago
I don't think gambling is the right analogy at all. I do think it can be addictive, but there are many things that are addictive that aren't gambling. I think a better analogy is something like extreme sport, where people can get
688.
by nl 8mo ago
"the post-training performance gains in Qwen3.5 primarily stem from our extensive scaling of virtually all RL tasks and environments we could conceive." I don't think anyone is surprised by this, but I think it's interes
689.
by nl 8mo ago
Are there any benchmarks (or even vibes!) about the token/second one can expect with this strategy?
690.
by nl 8mo ago
Do you have any references for that? AFAIK Anthropic won't let projects use the Claude Code subscription feature, but actually push those projects to the Claude Code API instead.

More ›