ac29

The data published in their S1 says they took a $693M operational loss in the first half of this year.

The other reasoning levels are almost always benchmarked, AA just doesnt show them in the default charts.

In my experience agents will reach for numpy for any non trivial math without being instructed

Its a paper launch, Argon took a few weeks from when it was "released" to when you could actually use it in agy

> For 5 units of 2 or 4 layer turnkey assembled boards, so long as you fit in the Grid Proto spec it's a little bit more expensive (not 10x, more like 1.3x)

Does that include the cost of parts? One of the issues I had with sourcing prototype PCBA's in the US is parts are much more expensive (not 10x, but also not 1.3x).

> Being mindful of wages is great, but eventually you have to accept that creating a job at a less than ideal wage is better than not creating any job at all.

For the employee, maybe. For the employer, it depends on how much time/money you have to put into training and what turnover is. Paying non-living wages leads to high(er) turnover because people will leave for anything better.

Yeah, I spent the last week with Deepseek Flash 4.1-1017, Kimi 3, and GLM 5.3 (and 5.3-flash). None of them were literally useless but also none of them were capable of making progress on the main project I am working on where Opus, Sol, and even Gemini Flash have all done very good work. Benchmarks might have you believe open models are only barely behind the big US labs in intelligence, but it feels much worse in practice.

> How long until China has its own chips?

There are dozens of Chinese domestic AI accelerators: Huawei, Alibaba, and Baidu all have chips, there are a bunch of smaller startups too. The chinese mainland is behind on leading edge fabrication though, SMIC only does ~7nm class since they don't have EUV.

Yeah at the point that Opus has read enough context to hand off a detailed plan, implementing the plan may not be that many more tokens.

Also having spent the weekend with the supposedly great Deepseek Flash 4.1 1017, I wonder if the people singing its praises have used recent frontier models. DS Flash feels like maybe Sonnet 4.5-4.6. Absolutely falls apart at things Opus/Sol currently excel at with low reasoning, and even has basic problems with failing tool calls, which hasnt been an issue with Anthropic/OpenAI in forever.

Could be a configuration issue with the provider I tried but honestly, I didnt get much better out of GLM5.3-flash. GLM5.3 and Kimi3 are much better but since most inference providers cant take a loss they are more expensive than just using Opus/Sol on subs, which are almost certainly currently a loss leader for their respective API offeringss.

> Huge, generational mistake for the USA to be leaning so hard into fossil fuels right now and actively trying to prevent buildout of solar and wind power.

If the US is trying that, they are failing pretty hard - 88% of new power in the US in 2025 was renewables, and 2026 is estimated to be above 90%

Openrouter does >600 trillion tokens a month and the top models shift on a daily basis. A lot of people are paying attention to new releases and even pre-releases (which tend to be free)

I'd rank Sol 6.1 only slightly behind Opus 5.5 in intelligence and both significantly ahead of anything else I've used (though cheaper/faster models are totally fine for a lot of tasks). My only real annoyance is 6.1 Sol is one of the slowest models I've used, which does make Opus 5.5 more pleasant to use.

There is a one click update in the WebUI now, no need to flash random bins

No, since that's a moving target. But the large GLM models are likely in a similar model weight class to the Opus family, just a few months behind

GLM 5.3 is roughly Opus class and is much, much cheaper

> I think we have the moral imperative to document all "out of distribution" information somewhere

This is more or less why writing was invented thousands of years ago, the amount of knowledge exceeds human memory

Have used decision models the same way: Had LLM generate/tune questions for Clef to label/extract emails with {useful work thing} content

Too early to say how useful this is versus existing techniques, the cool thing is this is the first thing I've found local models genuinely useful for. I think I was able to process 50,000 email questions per hour or so with Clef flash on Strix Halo with a custom engine. A modern gaming GPU would likely be much faster

> which is weird when Anthropic have supposedly claimed two profitable quarters in 2026

Anthropic claimed to be profitable if you dont count some of their largest expenses (like model training). Not the sort of profitable where revenue actually exceeds total expenses

Not a new revelation, but I think its funny that there is a inflate_signal_strength_bool setting and many carriers set it to true

Compute and payroll

Yeah, the small, useful bits of code I've made this year at work could have been contracted out in years past (we're a tiny non-tech co). But in addition to expense, there's the hassle of coordinating it, QA, etc. Really dont need all that for a project that connects a couple APIs with a minimal UI to help automate something previously done by hand.

You can run this model Clef, and many others, in llama.cpp

> Nearly for all submissions, there is one unquantified, unproven comment at the top saying it's AI, with just emotion and no proof

nearly all of the commits in the repo are co-authored by claude

If you're actually running 2 sessions 16 hours a day, I would hope you are getting $180 of value. (also, maybe work less)

> How many programmers really exist out there? Surely it can’t support the weight of investment that exists in AI already. It’s just such a small pool of the human race

I dont consider myself a programmer but use LLMs almost exclusively for coding.

The number of people able to create useful software today is much much larger than it used to be and arguably a minority of these people are/were "programmers"

> Sign in with ChatGPT: how is this not major news? They're launching with partnerships with a bunch of harnesses so you can use your subscription outside of Codex.

Maybe the sign on process is new but OpenAI has explicitly allowed subscriptions to be used in 3rd party apps at least as long as I've been a subscriber (a few months)

I dont think the post you are replying to was saying gambling on lotteries is rational. They are just saying the best possible outcome of buying a lottery ticket is much, much higher than buying an index fund (and the average, median, and worst outcomes are much worse)

I use a tiling wm, but the split view is more compact because it doesnt have two (vertical) tab bars. Useful on a cramped laptop, less so on a large screen

I've eaten wine grapes and while they taste... winey? I also quite like them

Not $518B in the next year, they plan "to spend $518 billion on cloud, computing and infrastructure obligations in coming years" (number of years unspecified)