blovescoffee

I don't know of a single zero day found on a number of tokens that fits inside a subscription plan. I'd be happy to be wrong.

1. it's not cheap to run glm-5.2 so not just anyone can do it 2. just because you haven't heard of attacks doesn't mean they haven't happened 3. this attack in the article was performed by a prerelease model which presumably benchmarks a bit above Sol which benchmarks above glm-5.2

We went from gpt 3 to models discovering and chaining their own zero days in a couple years. I'm not sure what else "takeoff" could possibly look like?

Huggingface literally reported the outage separately and did not know who caused it at first.

Roughly, you subscribe and listen to an observable. Incrementals are more like a cache across some DAG of computation + state that lets you optimize by only recomputing what needs to be recomputed.

There's a really good talk from Ron Minsky here: https://www.janestreet.com/tech-talks/seven-implementations-...

because you're looking up a result not solving for it

They emailed current paying users of the api (or at least that’s how I got updated).

Excited for the deepseek release this week (or at least they announced they'd release this week). Hopefully they also push even closer to SOTA.

"Current GPT/Claude/Gemini" is not a meaningful statement about perf. There's many different models from each of those providers and there's a considerable gap between the best of anthropic and open ai compared to gemini.

Benchmarks have GLM 5.2 somewhere underneath Sol and Fable and closer to now last-gen openai and anthropic models.

Some photos in the readme would be really great here.

I'd rather use

Is different than

quite literally unusable

I used a lot of Anki for spanish on top of classes. I'd say (anecdotally) it can bump you up a half to full letter in the A-C system depending on where you start. I learned far more living in Mexico though. You really need to talk to real humans to learn another language. If you don't, you can have a decent rating in A-C system but not actually converse that well.

Why would you bet against it being solved by LLMs? Isn't this very post proof that LLMs in an agentic harness are capable of doing real math? If you just keep cranking away at the tokens I don't see a principled argument against that leading to more solutions to unsolved math, even the hardest problems.

I work on a site with millions of users and we use bun...

I guess I would read this as “copper” as a primary requirement / necessity and some other metal as a secondary requirement. Are you just being pedantic?

Similar to running arch Linux. Many people do need to. But many people just like tinkering. Tinkering can lead to positive outcomes but it’s usually not “doing work”.

Pi is so “unbloated” that it’s extra effort to use. You can decide how much work to put into it. I get the trade off. But this is a big jump from CC. I’d recommend some middle ground like opencode.

Companies like Mercor sell data from human experts

I suppose it depends on what you count as an ingredient, even still it's greater than 7 thousand. I'm not sure what you mean by it's "your specialization" so I'm happy to hear what you could elaborate on, but I'm a bit stumped. Almost any given state in southern Mexico or any of the tropical latin American countries has thousands of ingredients from native herbs, leafy greens, chiles, spices, bugs, sea life, mammals, snakes and lizards, birds, ferments, mushrooms and fungus, tree barks, seeds, milks, any number of vegetables and fruits. Even an animal has non "meat" ingredients like blood, rennet and bile, bone marrow, do you count brains and intestines and offal here? I can't imagine where you're getting this less than 7 thousand number nor your confidence about it to be honest.

The timelines you’re quoting are too short but “ai is just git clone and sed” is not a good faith argument in 2026.

There are only a “few thousand ingredients” seems like orders of magnitude off

Only bc we’re on a nerd site, miso doesn’t involve lacto fermentation in any significant way. And I guess you can pickle cherries in any number of ways that aren’t lacto fermentation.

Wikipedia has demonstrably been ravaged by anti-semitism. Feel free to ignore any of the notes here about Israel / anti-Israeli sentiment which I understand is not clear cut. There's demonstrably antisemitic coalitions editing wikipedia en-masse.

https://www.adl.org/resources/report/editing-hate-how-anti-i...

Naive question, could agents help speed up building code for ROCm parity with CUDA? Outside of code, what are the bottlenecks for reaching parity?

There's definitely a few reasons but one of them is that you have to ask the GPU to do ~60x less work when you render 60x less frames

Do you really recommend people travel internationally in 2026 without a cellphone? I’m kind of bewildered by this suggestion. As someone who has to go between LATAM and US frequently, I have no choice but to bring my cellphone.

I'm working on this. It works pretty well. The main issue I'm working out right now (which has proven very difficult) is the auto-placing and auto-routing on a multi-layer pcb.

Okay read anything about David Ogilvy

customer acquisition

Not quite, compression enables you to simulate / represent / encode x data with less than x memory.

Source?