nicce

> If you were to take away my work's $500 budget for Opus 5.5 and replace it with unlimited use of GLM-5.3, it would be a huge downgrade in my opinion.

I am perfectly able to do the tasks with own custom harness and unlimited local Qwen 3.8 27B with 150t/s. Harness needs search capability and more fine-tuning for the job.

You need less words to describe the intent with Astra/Opus 5.5, and it automatically usually observes the best practice that you don't need to explicitly tell, but with only a slight more effort you can get the job done. I have even made security audits with Qwen 3.8 27B against projects fully written with Claude Opus and it is a bit mind-boggling, but there have been tens of vulnerabilities, identified by this model alone. This excludes the issues found when guiding more myself.

I think people were not satisfied with that one in S-1. It excludes this year completely and run-rate revenue and net revenue retention not verified, which is kinda big thing. They claim 900% NRR.

That is something obvious to add on relevant public materials.

> I wouldn’t trust models from China enough to use them in important work unless they were open.

One issue is that in the end, open models are still comparable to "freeware" from the old world. You have a black boxed binary to takes input and produces output. On typical "freeware", you could still reverse-engineer it and have have confidence to map assembly to actions it takes. But with these open models, not so much. So, we still can't put too much trust to these models, even if they are open. There can be some backdoor that is impossible currently to identify based on reading the model weight numbers. We can try to post-train models and reduce the likelihood of some "known" stuff, but we don't know what unknown there is. Interesting times ahead.

Bonsai 2 has a bit different architecture as far I understand and speculative decoding is not so great yet. Before that point, original model is faster if you have the memory. Would be nice to see the speed with proper speculative decoder.

They keep doing anti-competitive things over and over again, and the end users seems to refuse to see it. They do whatever they can to get vendor-lock in.

If using GrapheneOS, you need to give permission to write/install packages.

Yeah. Installed Signal as raw apk in GrapheneOS and you just need to open it to install the update. It does not update automatically in background. But anyway, works.

Somehow that statement just screams that our model is not useful for security work

> How did they not become spacex/starlink

You need to go big first with the rockets.

If you have the rockets, everything else is easy later on. Collect millions from others to launch the satellites in the meantime.

> admit it because it meant more work and that their past existing work was shite.

And that is why it is now more usable. More affordable.

"Popular" usually means that supporting many use cases. Many use cases means a lot of code. Probably means increased complexity. Tailored hyperspecific firmware is much more usable now and maybe you can even formally verify it.

You can technically send then all the time and it is enough that they boot once. No need to know the time beforehand.

This looks great. Quite new one, hopefully will be maintained.

Thanks. I am already using this, it seems to be the best so far. But I feel something is still missing from there by the usability aspect. It also uses NextJS which I don't know what to think about...

I just wish there would be more pure JMAP email clients. But maybe soon!

> Personally, I do expect SpaceX to face consequences in the years to come.

Probably only and only when AOC or someone else from Bernie's camp gets to be president. Otherwise, unlikely.

I even forgot that it exist. I tried it many times but it was not good enough. Eventually I switched to Zed since that was better. Probably Zed killed it.

Some repo owners just review code that others submit, and don't write anything themselves. If the review process stays the same, does it matter? The fault is not generating the code with AI, rather, with faulty review process or not caring about the code, or not reading it. A different problem.

Based on translating their Chinese text on Firefox, it says "per card":

> The current service throughput of a single cluster is 155 trillion tokens per day, and the overall revenue is $38 per hour (at the lowest price) per card.

With Kagi translate (best), it says GPU

> A single cluster currently serves 155 trillion tokens per day, generating overall revenue of $38 per GPU per hour (at off-peak electricity rates).

https://zhuanlan.zhihu.com/p/2095560295930508953

It is not even Russia in this case. Just look what Meta and Trump and friends have been doing

What are the benefits of jj if in the end everything is still on git backend?

And probably OLED or something close to that

Without an account, you can’t even properly use your freezer

The main issue is that the attention economy and social pressure are so hard to break. People keep using WhatsApp because it is convenient and everyone is there. There are no other reasons.

The best games when you were around 10-12 years old.

> Or you build a cartel of apirin sellers to do some price fixing.

They are really trying to do that by getting that regulatory capture

> 1) If a firm spends $99k on compute to replace a $100k engineer and produces 5x the output, how is that not a 5x increase in aggregate labor productivity?

I think they are saying that with 99k pure compute you are able replace 100k engineer with equal output. Or that is the goal of the AI companies. It is worse for them if you get more productivity.

There is named entity in the end, and "we" statement. But yeah, I was speaking in general, not so specifically about this page. It is better than most who do this.

There is nothing I dislike more than people using AI to generate tens of pages in text and then expect people to read them, when they did not read them themselves.