storus

13 ms·

271.
by storus 7mo ago
Weren't 512GB models selling like hot cakes to the complete surprise of Apple? Wait time was up to 3 months last time I checked. Glad I got mine last October.
272.
by storus 8mo ago
I am wondering how much did war in Ukraine contribute to it? On one hand, EU is pushing green agenda, on the other hand there are daily oil storage facility explosions evaporating weeks of oil consumption in a few hours.
273.
by storus 8mo ago
Why are some people able to bear suffering whereas others go bonkers? Or what if the only source of happiness of some of those crazy people is domination of other people and exclusivity of social hierarchies? How would AI fix that?
274.
by storus 8mo ago
> to decide halfway through following my detailed instructions that it would be "simpler" to just... not do what I asked That's likely coming from the 3:1 ratio of linear to quadratic attention usage. The latest DeepSeek a
275.
by storus 8mo ago
More likely some high ranking party member's nepobaby from Gemini sniffed success with Qwen and the original folks just walked away as their reward disappeared.
276.
by storus 8mo ago
The article directly mentions it; see the section 'Leaving wasn't your choice'.
277.
by storus 8mo ago
Developing a business product and monetizing it doesn't require sacking the research group that created it and would have developed it further.
278.
by storus 8mo ago
Happening to Skoda in Volkswagen - German managers were first forcing Skoda to have dumb design to not threaten VW/Audi, that didn't work, Skoda became the only profitable part of VW, now they are imposing 25% cost reduction at Sk
279.
by storus 8mo ago
This feels like a typical sociopathic corporate scenario with mushroom management - let a bunch of nerds develop something new/exciting outside mainstream corporate culture, then once it becomes good enough jump in, cut them off and ha
280.
by storus 8mo ago
I delegate to agents what I hate doing, e.g. when creating a SaaS web app, the last thing I want to waste my time on is the landing page with about/pricing/login and Stripe integration frontend/backend - I'll just tell C
281.
by storus 8mo ago
These lessons get obliterated with every new LLM generation. Like how LangChain started on stupid models with small context, creating some crazy architecture around it to bypass their limitations that got completely obliterated when GPT-3.5
282.
by storus 8mo ago
Any ICE passing Frankfurt is a guaranteed 30 minute delay. Too easy to win.
283.
by storus 8mo ago
In my case no kernel extensions. Every few days the computer freezes with a beachball and/or screen starts falling apart in large blocks randomly shuffling around the screen and I need to power it off. The first logicboard also added a
284.
by storus 8mo ago
My 128GB RAM M3 Max had logic board replaced 3x and I am still getting beachball alongside screen falling apart in blocks... There is something wrong with their firmware, especially when you are switching between multiple users often.
285.
by storus 8mo ago
4x faster is about token prefill, i.e. the time to first token. It should be on par with DGX Spark there while being slightly faster than M4 for token generation. I.e. when you have long context, you don't need to wait 15 minutes, only
286.
by storus 8mo ago
As long as python runs all the models, the best language for agents is likely Python as it allows e.g. auto-fine-tuning of (local) LLMs for self-improving agents without the need to change the programming language. Use Pydantic if you care
287.
by storus 8mo ago
There is a huge supply chain surplus of notch displays as nobody wants them so I guess they decided that "real open source" folks don't care about design and bought them for pennies.
288.
by storus 8mo ago
Local inference might a better bet for you.
289.
by storus 8mo ago
Because you need to understand the output of coding agents when there is an issue or even a decision to be made. Even Opus-4.6 gets into rough states where it spews out garbage or suboptimal code and juniors might not catch it. Reading some
290.
by storus 8mo ago
Maker movement was a great success, but in China, not in the US.
291.
by storus 8mo ago
Chat.com is ChatGPT so likely that's why he called it Chat.
292.
by storus 8mo ago
I'd say using them as draft models for some strong AR model, speeding it up 3x. Diffusion generates a bunch of tokens extremely fast, those can be then passed over to an AR model to accept/reject instead of generating them.
293.
by storus 8mo ago
I think that's a natural outcome of a model where sociopaths climb to the top, with a layer of sycophants beneath them that shield normal workers from perceiving the amount of depravity going on at the top which would make them unable
294.
by storus 8mo ago
This is really YOLOing as the original author doesn't know Rust well so what happens if they hit some complex production issue LLM aren't aware of? Hiring an expensive consultant to fix that until the next LLM iteration?
295.
by storus 8mo ago
Dunno, that probably doesn't hold for webapps with backend as they are typically complete garbage and LLMs (even local ones) would give you about the same result but in 1 hour.
296.
by storus 8mo ago
That won't work for immediate relief; placebo is studied statistically as a cumulative change over a period of time. Not when I put Gunnars over my sore eyes from computing too much and get immediate relief. Frankly, I find this discus
297.
by storus 8mo ago
My Gunnars give me immediate eye relief after a day of work looking at a display. Are you sure it's a scam?
298.
by storus 8mo ago
RL suffers both from sparsity in a huge amount of dimensions as well as convergence brittleness when it's extremely difficult to get it to converge.
299.
by storus 8mo ago
Didn't thinking tokens resolve the most problematic part of autoregressive models (the first few tokens set the constraints the model can't overcome later) and give it a massive advantage compared to diffusion models by showing th
300.
by storus 8mo ago
I'll bite: Who is a wellbeing expert? What is the ground for calling some people wellbeing experts? How does one earn it? How general is their expertise? Would you trust your psychologist with everything they say? Or even a group of ps

More ›