sroussey

11 ms·

271.
by sroussey 6mo ago
Actually, they published a year ago. Recent was being on official Google blog. https://arxiv.org/abs/2504.19874 https://research.google/blog/turboquant-redefining-ai-effici...
272.
by sroussey 6mo ago
Don’t forget http pipelining!
273.
by sroussey 6mo ago
I tried /debug as the only input, hoping CC wouldn’t shit the bed and give me some data. Heck, just saying “hello” causes Claude Code to fail. I’m thinking of doing a charge back, and creating a new account. Others don’t seem to have t
274.
by sroussey 6mo ago
Thinking Machines Lab uses batch invariant kernels, btw.
275.
by sroussey 6mo ago
Super interesting! > - People have successfully used TurboQuant to quantize model weights (TQ3_4S), not just the context KV, to achieve smaller sizes than Q4 (~3.5 bpw) with much better PPL and faster decoding. Where can I find more info
276.
by sroussey 6mo ago
Have a look at https://github.com/anthropics/claude-code/issues/54497 I can’t use Claude Code online at all
277.
Anthropic fails worse than Githubs(github.com)
5 points by sroussey 6mo ago|2 comments
278.
by sroussey 6mo ago
What’s the point of a Max plan if you can’t use it? At least git still works without GitHub. (I guess normal chat works so if paying for Claude Code, there is a conciliation prize?) The funny part is that the support AI is terrible! CC repo
279.
by sroussey 6mo ago
Enter Cursor
280.
by sroussey 6mo ago
I think the Venn diagram of people letting LLMs have complete control of their database AND having good backups, will have no overlap. The people that would benefit or not the people that have backups.
281.
by sroussey 6mo ago
I thought these days thinking tokens sent my the model (as opposed to used internally) were just for the users benefit. When you send the convo back you have to strip the thinking stuff for next turn. Or is that just local models?
282.
by sroussey 6mo ago
Oh, absolutely. Though changes in how the model is used is imminently more fixable than the model itself.
283.
by sroussey 6mo ago
None of these problems equate to degrading model performance. Completely different team. Degraded CC harness, sure.
284.
by sroussey 6mo ago
> The standard price for a GB of egress from a cloud provider is 10x what you pay racking a server in a normal data center. Oh, that’s too kind. More like 100x to 1000x. Raw bandwidth is cheap.
285.
by sroussey 6mo ago
Many cloud vendors have you pay through the nose for IOPS and bandwidth. Edit: I posted this before reading, and these two are the same he points out.
286.
by sroussey 6mo ago
I’ve seen ai coding agents spin out and create 1_000 line changesets that I have to stop before they are 10_000. And then I look at the problem and change one line instead.
287.
by sroussey 6mo ago
How to you get CC to connect to your dev container? I have the CC app but it’s kinda useless as I’m not have it barebacking my system, so I’m left with the cli and vs code extension.
288.
by sroussey 6mo ago
At the moment being important. Previous day it was higher at 38.8% but only for that 15min. As a percent of the whole day it’s much much lower. (California goes to 110% in the middle of the by comparison). It’s a good upward trend, I’m conf
289.
by sroussey 6mo ago
What pollution? Last I was using data centers directly there was no water use, though I know now that many use water for cooling and don’t bother with a closed system because water is cheaper than the power. (Exception being Elon and gas tu
290.
by sroussey 6mo ago
CC has such egregious API subsidies that it’s hard to not to leverage it unless the license tells an enterprise otherwise. Love the subsidized pricing while it lasts.
291.
by sroussey 6mo ago
How do you hit that limit? I’m never close to it.
292.
by sroussey 6mo ago
All the money they are burning is for grok. And it is not decades ahead.
293.
by sroussey 6mo ago
No, the burn up in the atmosphere. Burning metals being added to the oxygen you breathe.
294.
by sroussey 6mo ago
So is pace travel. Then rockets are not green!
295.
by sroussey 6mo ago
If I stop paying for Cursor, will they threaten to sue like Twitter does?
296.
by sroussey 6mo ago
SpaceX was, but SpaceTwitter is not. xAI is hoovering all the money out of SpaceX.
297.
by sroussey 6mo ago
In a way, SOLID is premature optimization. You are optimizing abstractions before knowing how the code is used in practice. Lots of code will be written and never changed again, but a minority will see changes quite a bit. Concentrate there
298.
by sroussey 6mo ago
Except the iteration on it. And people we aghast at the cost. But one product. I don't know man, I think they became chicken of anything grand. It is not like it was a $35000 product. If the Vision Pro was the Lisa, where is the Vision
299.
by sroussey 6mo ago
The Apple Lisa was the first GUI computer Apple made. Starting price $9995 (or $35,000 in today’s dollars). Yes, Apple has gone down market these days, but their history is really premium. Or they start premium and then move down market lik
300.
by sroussey 6mo ago
Or the other likely version: prevent commercialization. No source means that someone can’t make a fork, put on a new domain, run ads and charge money for his work.

More ›