2001zhaozhao

15 ms·

331.
by 2001zhaozhao 9mo ago
At least the AI teams aren't politically competing against each other unlike human teams. (Or are they?)
332.
by 2001zhaozhao 10mo ago
I think "highly processed foods are bad" is best seen as a general rule and no more than that. However, it is a good general rule and following it is probably the easiest way for people to eat healthy. In general, the more proce
333.
by 2001zhaozhao 10mo ago
> - Streaming game primitives instead of pixels, which should be much lighter - Server-side rendering makes cheating basically impossible This doesn't work in 3D. Unless you have the server do the work of the GPU and compute occlusi
334.
by 2001zhaozhao 10mo ago
Wayland just has the best multi display support hands down. (I am using Cosmic DE but KDE is similar) Windows: - Per-display scaling can be set to multiple of 25% (you can set more precise scaling ONLY if you apply it to all displays) - Win
335.
by 2001zhaozhao 10mo ago
It really is cursed to be spending hundreds of watts of power in a datacenter somewhere to make a laptop run slightly faster.
336.
by 2001zhaozhao 10mo ago
In short it's striking a careful balance between giving game mods the flexibility to make whatever they want, and not giving them too much low-level access that it leads to incompatibilities with other mods. In my case, I am not lettin
337.
by 2001zhaozhao 10mo ago
It's a lot more expensive than normal, $2.25/2.75 I think. Though their subscription is a lot cheaper.
338.
by 2001zhaozhao 10mo ago
If I were predicting the bubble to burst and API prices to go up in the future, wouldn't it be much better to use (abuse) the cheap API pricing now and then buy some discount AI hardware that everyone's dumping on the market once
339.
by 2001zhaozhao 10mo ago
What we really need is well-built open source software base with flexible and well-documented plugin SDKs where people can just vibecode their own extensions on top of them. The software needs to be designed from scratch expressly for the p
340.
by 2001zhaozhao 10mo ago
Cerebras is serving GLM4.6 at 1000 tokens/s right now. They're probably likely to upgrade to this model. I really wonder if GLM 4.7 or models a few generations from now will be able to function effectively in simulated software de
341.
by 2001zhaozhao 10mo ago
Under current prices buying hardware just to run local models is not worth it EVER, unless you already need the hardware for other reasons or you somehow value having no one else be able to possibly see your AI usage. Let's be generous