alecco

9 ms·

151.
by alecco 5mo ago
Thanks. Discussion https://news.ycombinator.com/item?id=48054182
152.
by alecco 5mo ago
Or disable algif_aead module as in https://news.ycombinator.com/item?id=47957409
153.
by alecco 5mo ago
Are Googlers themselves happy using Gemini coding agent instead of Claude Code or Codex? (no snark, I'm really asking)
154.
by alecco 5mo ago
Yes, they are loaded, that's the point. It's a rhetorical method to (hopefully) make people consider that point of view.
155.
by alecco 5mo ago
I hope you enjoyed the hormonal reward you got when you copy-pasted that amazingly deep proverb. Let's expand on it a bit. Is the world a mirror to street salespeople? Is the world a mirror to stalkers? To sociopaths?
156.
by alecco 5mo ago
> One of the things I like about this is that OP is giving people genuine compliments without any particular agenda. IMHO that is as fake as a car salesman. Mature, cultured people will say thank you and think " what a nuisance &quo
157.
by alecco 5mo ago
My grandparents just bumped their volunteering from weekends to weekdays. Then my Boomer parents switched to leisure activities and travel (they stopped volunteering when they retired). I prefer my grandparent's retirement, but now tha
158.
by alecco 5mo ago
Bait question titles piss me off. But there you go, I removed it.
159.
Is China decoupling on food?(ft.com)
3 points by alecco 5mo ago|3 comments
160.
by alecco 5mo ago
https://archive.ph/GbpPS
161.
by alecco 6mo ago
> If it was actually productive They are extremely productive if you use them right. To the point it worries me how clever these pseudo-AI models can get in the next year.
162.
by alecco 6mo ago
UK: hold my beer...
163.
by alecco 6mo ago
I prefer distributed power, not a democratic system as it's often abused. If large models become a +100x productivity multiplier they can charge crazy money for access. So rich/money people will dominate the world in no time. Tod
164.
by alecco 6mo ago
Guys, did you bother checking the poster's profile? https://xcancel.com/lifeof_jer . SEE THE TWEET BELOW. Smells like a ragebait post to me. Also search online for his alleged "PocketOS" company with software
165.
DeepSeek drops input cache price to 1/10th(xcancel.com)
6 points by alecco 6mo ago|1 comments
166.
by alecco 6mo ago
Long chats are mostly cache. For coding agents this is like >90%.
167.
by alecco 6mo ago
CoRecursive had a really good episode about this last August: "Coding in the Red-Queen Era" https://corecursive.com/red-queen-coding/
168.
by alecco 6mo ago
Cool story, SEO bro.
169.
by alecco 6mo ago
I meant it as a simple to understand parallel. Absolutely deep reading and thought is much better than Wikipedia or an LLM chat.
170.
by alecco 6mo ago
Speak for yourself. I now dare to code much harder problems and learning is bliss. No more having to sit down to dig needle-in-haystack through horrible documentation or random Stack Overflow posts. LLMs are a magnificent tool if you use th
171.
by alecco 6mo ago
> DeepSeek is famous for publishing everything. They might take a bit to publish source code but it's almost always there. they-might-take-a-bit-to-publish
172.
by alecco 6mo ago
>> Truly open source coming from China. > Open weight ! They clearly were implying it's not open source.
173.
by alecco 6mo ago
X is full of "open weights!" corrections as a dog whistle by the anti-China crowd. And they are right about models from the Chinese Big Tech, but completely wrong about DeepSeek.
174.
by alecco 6mo ago
> These are open weights, not open source. Did you even read my comment?
175.
by alecco 6mo ago
If 5090 has 32GB, and let's say somehow a 1-bit quantization is possible and you don't need more VRAM for anything else (forget KV cache etc), it would be able to fit a 256B 1-bit model. Just to picture it in extremes how unlikel
176.
by alecco 6mo ago
The paper is about previously unknown ways coffee affects the body.
177.
by alecco 6mo ago
Please don't slander the most open AI company in the world. Even more open than some non-profit labs from universities. DeepSeek is famous for publishing everything . They might take a bit to publish source code but it's almost a
178.
OpenAI's Codex grew 33% in the last 2 weeks (active users 3M –> 4M)(xcancel.com)
2 points by alecco 6mo ago|1 comments
179.
by alecco 6mo ago
Related interesting find on Qwen. "Qwen's base models live in a very exam-heavy basin - distinct from other base models like llama/gemma. Shown below are the embeddings from randomly sampled rollouts from ambiguous initial wo
180.
by alecco 6mo ago
Yes, I was thinking that. But it could as well be the other way around. Using the pretrained 4.7 (1T?) to speed up ~70% Mythos (10T?) pretraining. It's just speculative decoding but for training. If they did at this scale it's qui

More ›