naasking

10 ms·

181.
by naasking 6mo ago
Quantization can introduce these issues, and Gemma 4 also had issues because the prompt tokens that Gemma used was new and not well supported yet.
182.
by naasking 6mo ago
MoE isn't inherently better, but I do think it's still an under explored space. When your sparse model can do 5 runs on the same prompt in the same time as a dense model takes to generate one, there opens up all sorts of interesti
183.
by naasking 6mo ago
The noise we make is hurting humans too.
184.
by naasking 6mo ago
> When I mentor new devs, I explain to them how I use git. Now imagine not needing to do that. > I don't see how. Safer programming language address a clear problem in C Being productive in C means training and experience in avoi
185.
by naasking 6mo ago
> then it's irrelevant to those of us who have been lucky to find and learn sensible UXs Only if you're a solo dev that doesn't work on a team or have to mentor new devs that haven't developed good intuitions for this
186.
by naasking 6mo ago
> Then I went back to git (it's been 6 months now) and I haven't had a single case of "this is so painful, I wish something better existed". The core issues are: how long did it take you to get there, how many lucky d
187.
by naasking 6mo ago
> A missing specification in the proof of lean-zip, a lean component, is a real problem to the philosophy and practice of software verification. Every time someone makes this point, I feel obliged to point out that all alternatives to so
188.
by naasking 6mo ago
ROCm is so annoying (buggy, fiddly dependencies, limited hardware support) that TinyGrad built its own compiler and toolchain that targets the hardware directly. And it has broader device support than ROCm, which primarily seems focused on
189.
by naasking 6mo ago
CLAUDE_CODE_AUTO_COMPACT_WINDOW=400000
190.
by naasking 6mo ago
> If all you have is a spade, that is _not_ evidence that spades are good for excavating an entire hill. If you have an automated spade, that's still often better for excavating that hill than you using a shovel by hand.
191.
by naasking 6mo ago
> My proof-in-pudding test is still the fact that we haven't seen gigantic mass firings at tech companies Jevon's paradox.
192.
by naasking 6mo ago
I'm sure if they could, they would, sure, as would any business. That's where competition enters the equation. They can't do it because their competitors would undercut them by requiring no such conditions.
193.
by naasking 6mo ago
> I don't see why Facebook would get money because they helped people connect to each other who ended up developing a cancer cure, but they definitely should be held accountable for enabling a genocide. Why? What does it even mean t
194.
by naasking 6mo ago
Sure they would, just like people would use the bad PR to smear OpenAI if someone did something bad with knowledge their model created. The situation is totally symmetrical and fair as it is, and my point is that expecting them to liable is
195.
by naasking 6mo ago
> 1. LLMs don't just provide knowledge, they provide recommendations, advice, and instructions. That's knowledge. > 2. OpenAI very much feels that they should profit from the results of people using their tools. Even in heal
196.
by naasking 6mo ago
If an OpenAI model helped someone create a cancer cure, they wouldn't see a dime from that beneficial act. So why should they be liable if someone does something harmful with the model?
197.
by naasking 6mo ago
Making knowledge illegal is a dangerous precedent. Actions should be illegal, not knowledge. Don't outlaw knowing how to make neurotoxic agents, outlaw actually trying to make them. As for OpenAI immunity, I'm not sure I see the p
198.
by naasking 6mo ago
> Based on what? A lot of this is vibes and FOMO; just like any economic bubble. You're in a bubble. https://www.helpnetsecurity.com/2026/04/07/google-llm-conten...
199.
by naasking 6mo ago
> The best you'll see is an improvised post-hoc rationalization story. Funny, because "post-hoc rationalization" is how many neuroscientists think humans operate. That LLMs are stochastic inference engines is obvious by co
200.
by naasking 6mo ago
I think many humans engage in metacognitive reasoning, and that this might not be strongly represented in training data so it probably isn't common to LLMs yet. They can still do it when prompted though.
201.
by naasking 6mo ago
> Conversely: in humans, intelligence is inversely correlated with crime. Inversely correlated with crime that's caught and successfully prosecuted , you mean, because that's what makes up the stats on crime. I think people to
202.
by naasking 6mo ago
I'm curious if frontier labs use any forms of compression on their models to improve performance. The small % drop of Q8 or FP8 would still put it ahead of Opus, but should double token throughput. Maybe then interactive use would feel
203.
by naasking 6mo ago
I used GLM5 quite a bit, and I'd say it was maybe on par with Sonnet for most simple to medium tasks. Definitely not Opus though. Didn't test super long context tasks, and that's where I would expect it to break down. A recen
204.
by naasking 6mo ago
Very interesting. I run Claude Code in VS Code, and unfortunately there doesn't seem to be an equivalent to "cli.js", it's all bundled into the "claude.exe" I've found under the VS code extensions folder (
205.
by naasking 6mo ago
They're a business. The alternative to keep costs in check would to ask you for more money, and you'd likely be even more upset with that.
206.
by naasking 6mo ago
It's interesting that LLMs improve skills, especially on harder problems, just by practicing them. That's effectively what's going on.
207.
by naasking 6mo ago
> I only ask because I've been running local models (using Ollama) on my RX 7900 XTX for the last year and a half or so and haven't had a single problem that was ROCm specific that I can think of. It's probably using the V
208.
by naasking 6mo ago
Routing in a MoE model might fit.
209.
by naasking 6mo ago
It's just an example where it fits perfectly, and it's exactly what something like Alexa or Google home needs for low power machine learning, eg. when sitting idle it needs to consume as little power as possible while waiting for
210.
by naasking 6mo ago
Small models aren't entirely useless, and the NPU can run LLMs up to around 8B parameters from what I've seen. So one way they could be useful: Qwen3 text to speech models are all under 2B parameters, and Open AI's whisper-sm

More ›