naasking

13 ms·

241.
by naasking 7mo ago
Reich's lab actually found evidence of meaningful genetic changes that improved intelligence over the past 10,000 years, but not so much prior to that: https://www.biorxiv.org/content/10.1101/2024.09.14.613021
242.
by naasking 7mo ago
> You don't get that many iterations in the real world though True, for iterations between the same two players, but humans evolved the ability to communicate and so can share the results of past interactions through a network with
243.
by naasking 7mo ago
> Algorithms do not possess ethics nor morality[0] and therefore cannot engage in Machiavellianism[1]. Conjecture. There are plenty of ethical frameworks grounded in pure logic (Kant), or game theory (morality as evolved co-operation). T
244.
by naasking 7mo ago
I don't think old prompts would become useless. A few studies have shown that prompt crafting is important because LLMs often misidentify the user's intent . Presumably an AI that is learning continuously will simply get better a
245.
by naasking 7mo ago
A single DGX Spark can service a whole department of mathematicians (or programmers), and you can cluster up to 4 of them them to fit very large models like GLM-5 and quants of Kimi K2.5. This is nearing frontier-level model size. I underst
246.
by naasking 7mo ago
> Atmospheric entry (if that's what you mean) is irrelevant. I think the OP meant that Earths magnetic field and atmosphere shields any terrestrial matter far more than than a bare asteroid that has no such protections, so it seems
247.
by naasking 7mo ago
> And we're sort of back to square 1. Specifications are smaller than the full code, just as high level code is smaller than the functionally equivalent assembly. As we ascend the abstraction ladder the amount of reading a human nee
248.
by naasking 7mo ago
You can easily run a quant of this on a DGX Spark though. Seems like a small investment if it meaningful improves Lean productivity.
249.
by naasking 7mo ago
Does a warrant ever expire? How long can they monitor you once the warrant is issued? Do they ever have to notify you or anyone else that you were being monitored and they found no criminal conduct? Don't you see the potential for abus
250.
by naasking 7mo ago
SDFs can be made fast: https://www.youtube.com/watch?v=il-TXbn5iMA
251.
by naasking 7mo ago
Here is the Veritasium video where they review ASML's EUV lithography in depth: https://www.youtube.com/watch?v=MiUHjLxm3V0 High thermal conductivity, inertness, and low absorption of the EUV wavelength used are the pr
252.
by naasking 7mo ago
Of course vibe coding is going to be a headache if you have very particular aesthetic constraints around both the code and UX, and you aren't capable of clearly and explicitly explaining those constraints (which is often hard to do f
253.
by naasking 7mo ago
That's not clear actually. It's been awhile since I watched the video on the ASML EUV system for etching wafers, but from my recollection, the UV light ablates away layers of wafers and so particles of silicon presumably contamina
254.
by naasking 7mo ago
> They'll be thinking "Of all possible things we could talk about, why this??" Sure, but the point is that they should just say that, and the person who raised the issue must then explain why it's worth their time.
255.
by naasking 7mo ago
True, but couldn't you just simulate it by enqueing a thunk/continuation?
256.
by naasking 7mo ago
I assume purity matters. It probably picks up a lot of contaminants during use, and chip manufacturing is pretty sensitive to contaminants.
257.
by naasking 7mo ago
It's a little weird that you seem to think that humans can't make technological progress without globalization.
258.
by naasking 7mo ago
Redundancy is not "hate", it's robustness against failure. Complex, interconnected supply chains are fragile. If you're building an online storage API, you would (I hope) consider it deeply irresponsible not to setup ple
259.
by naasking 7mo ago
How does GLM-OCR compare to Qwen 3 VL? I've had good experiences with Qwen for these purposes.
260.
by naasking 7mo ago
> And on the flip side I can easily see why not allowing email addresses to be used again is a reasonable security stance, email addresses are immutable and so limiting them only to one identity seems logical. If they aren't actuall
261.
by naasking 7mo ago
> I ended up forgetting to visit Lemmy or Pixelfed or <whatever> because I had 2-3 times when I opened up the app and saw the exact same content, giving me a feeling of "nothing is happening here" and thus, I didn't
262.
by naasking 7mo ago
> Model's are no longer revolutionary, they are evolutionary, and the evolution and per model-version difference is narrowing each release. We've definitely culled some low hanging fruit, but I think there's still a lot of
263.
by naasking 7mo ago
Sounds reasonable, but gains will go up. There is a ceiling somewhere, but we don't know where it is.
264.
by naasking 7mo ago
I don't think there is a new model, they recently released a new inference engine for CPUs for bitnet models: https://github.com/microsoft/BitNet/blob/main/src/README.md
265.
by naasking 7mo ago
Thanks for the link, the GSM8K result actually leads the pack in that table, but math is indeed underwhelming. Qwen 2.5 is in the lead, but bitnet isn't far behind and it takes 1/6th as much memory during inference, and was traine
266.
by naasking 7mo ago
I think the idea is to train a small, minimal LLM thinking model that can run on edge devices, but that has very little knowledge embedded in its weights, and so performs a sort of RAG to Encylopedia Britannica to ground answers to user que
267.
by naasking 7mo ago
I think the comparison is fair, strictly in the sense that both Rust and C++ are designed around extensible programming via a sort of subtyping (C++ classes, Rust traits), and similar resource management patterns (ownership, RAII), where Zi
268.
by naasking 7mo ago
It's a two year old base model that's only 3B parameters, trained on only 100B tokens. It's still a research project at this point.
269.
by naasking 7mo ago
I think the README [1] for the new CPU feature is of more interest, showing linear speedups with number of threads. Up to 73 tokens/sec with 8 threads (64 toks/s for their recommended Q6 quant): https://github.com/
270.
by naasking 7mo ago
What's unusual about it? It seems pretty standard to train small models to validate an approach, and then show that training scales with model size to 8B to 14B parameter models, which is what they did.

More ›