petra

4 ms·

31.
by petra 2mo ago
They can do 7nm. But they use multiple patterning(printing the same pattern multiple times to get to 7nm), which is expensive. So it's not comparable on cost to western single-patterning 7nm, and of course not to the leading edge on co
32.
by petra 2mo ago
AMD can let the SRAM be on a different chip. Maybe even something similar to their 3D cache. that could increase density to 20B[1]. They could also move from 6nm to 2nm. that would probably increase density by another 3x to 60B. Add a bunch
33.
by petra 2mo ago
I'm sure Europe currently uses the surveillance capabilities of those systems to prevent bad shit from happening. So maybe they prefer another solution to the sovereignty issue.
34.
by petra 2mo ago
Creation of nuclear reactors that are useless for terrorists could help
35.
by petra 2mo ago
Software engineers are in a decades long project of automating everyone and everything else, and getting the financial rewards out of that. So...
36.
by petra 2mo ago
So maybe your friend need a role reversal: A prompt guiding the llm to act as a consultant, guiding people in implementing a software project, asking questions, creating a shared understanding, limiting scope or creating milestones. That se
37.
by petra 2mo ago
They probably took our skills and reuse them. That's why skills do not matter anymore.
38.
by petra 2mo ago
It can follow instructions well. Just give it a spec in a formal, well defined language. runnable language and some reinforcement learning. It can also handle vague stuff. That's certainly more powerful than a regular computer language
39.
by petra 2mo ago
// makers want to make not ask ai to make, It probably changes from person to person, some people are just interested in seeing how their idea comes to life and maybe does something useful in the world.
40.
by petra 2mo ago
Also weird old hardware.
41.
by petra 2mo ago
But this guy will probably get hired because of this.
42.
by petra 2mo ago
And besides llm's memory requirements: -HBM has a lot of room to grow(bandwidth/power/density). And an HBM memory takes 3x the amount of wafers of regular DRAM. -HBM + better DRAM designs could increase random memory access s
43.
by petra 2mo ago
Given the limits of LLM's regarding context, why is zero dependencies the goal ? Shouldn't something like maximum reuse/abstraction, minimal written code, should be a better goal ? Possible leading to more llm-scalable system
44.
by petra 2mo ago
The other side of llm's is the need for rich feedback, often by simulations of various things. And memory latency is a big bottleneck there. So this chip nicely complements the llm's.
45.
by petra 2mo ago
LoL.maybe the question is not phrased in the best way, English as a second language and laziness, but it's a pretty decent question.
46.
by petra 2mo ago
Given LLM's is being smarter still a worthwhile to optimize with supplements?
47.
by petra 2mo ago
I wonder what perf. this cpu can get in bit/ternary inference, and whether the 512AVX units(or whats it's called on AMD), get throttled.
48.
by petra 2mo ago
You're right. I've used reader mode(the paper in the url bar in Firefox) to solve this.
49.
by petra 2mo ago
That's the current Netanyahu and right wing extremists government. Nothing good can be expected of them. Elections are going to be held soon. Hopefully this government will be replaced, and better things start happening.
50.
by petra 2mo ago
There's also digital probabilistic computing, with a processor working on AWS, and for the right workload ,claimed to be offering a very significant speed boost: https://www.eetimes.com/probabilistic-computing-is-alread
51.
by petra 2mo ago
We already drown in useful new things and especially knowledge. There's a limit to how fast we can adopt them.
52.
by petra 2mo ago
Llm's do have discipline and persistence, though.
53.
by petra 2mo ago
The ability to correct the plan/prompt.
54.
by petra 2mo ago
True, that's probably the biggest issue. But the only solution here is very expensive marketing, so...
55.
by petra 2mo ago
I agree. The market doesn't work well around this. You can ask chatgpt.not a great way. And when you ask it you the secure answers cost 3x more. And than require an installar. And some(Google) require a monthly subscription. And even a
56.
by petra 2mo ago
Consumers just don't care about security. It is what it is.
57.
by petra 2mo ago
No pricing, "call for demo". Not sure this is for personal use.
58.
by petra 2mo ago
Or maybe just a way to use chatgpt/Claude with files , and get accurate references to the page/paragraph in the source?
59.
by petra 2mo ago
I think the next popular web framework would be something that would be optimized against llm code generation. So that the end result would be secure. correct. scalable(and scaling should be done by an llm).
60.
by petra 2mo ago
That's an interesting question. What happens if we take the most abstract libraries in any given field - and: 1. Bound to the llm to only use those as building blocks. Does it affect his reasoning ? Will it think more abstractly ? 2. T

More ›