p1esk

8 ms·

61.
by p1esk 5mo ago
Do you mean Facebook to Yahoo?
62.
by p1esk 5mo ago
Why can't I want both earning 1B and do good things for the world? Unless his diary directly contradicts what he has said in public (like "I don't care about money"), I see zero moral issues here.
63.
by p1esk 5mo ago
I struggle to imagine how would something like that form, function, or become powerful enough to matter. Do you have any examples of such coalitions?
64.
by p1esk 5mo ago
Don’t you think a far bigger threat to democracy is that government allows this influence? And voters seem to be perfectly OK with both 250M campaign contributions and widespread lobbying?
65.
by p1esk 5mo ago
Other than researchers Are you saying researchers are less interested in quality of life than other people? If this was true, frontier labs wouldn’t need to offer 7 digit compensation packages to their researchers.
66.
by p1esk 5mo ago
We already have two such coalitions within Congress: Republicans and Democrats. How is that working out?
67.
by p1esk 5mo ago
The point is that modern CPUs are not as slow as most DL people think. Roughly 10x slower but with a lot more memory.
68.
by p1esk 5mo ago
Even if you could achieve that, there would still be rich people. Musk would still be a billionaire even if he had to pay 90% tax. Plus, many powerful people in government are not that rich.
69.
by p1esk 5mo ago
To do that you first need to become more powerful than those powerful people, right?
70.
by p1esk 5mo ago
We need to not have powerful people What does this even mean?
71.
by p1esk 5mo ago
No one in their right mind would use pure Python to do matrix multiplication. It’s like using a screwdriver to hammer nails into wood. But this discussion is even more bizarre than comparing a screwdriver to a hammer, it’s like comparing a
72.
by p1esk 5mo ago
Here’s a great book explaining microprogramming from ground up: https://www.amazon.com/Computation-Structures-Optical-Electr... Easy to find a free pdf
73.
by p1esk 5mo ago
Intel Xeon 6980P: 128 cores x 1024 FP16 FLOP/cycle/core x 3.2 GHz: 419 TFLOP/s
74.
by p1esk 5mo ago
Theoretical FP32 performance of AMD EPYC 9965 is double that of A100: 41.2 TFLOP/s vs 19.5 TFLOP/s
75.
by p1esk 5mo ago
A100 FP32 throughput “at its limit”: 19.5 TFLOP/s. AMD EPYC 9965 FP32 throughput “at its limit”: 41.2 TFLOP/s (192 cores x 64 FP32 FLOP/cycle/core x 3.35GHz).
76.
by p1esk 5mo ago
This statement makes zero sense
77.
by p1esk 5mo ago
Depends on TTFT and tokens per second you want.
78.
by p1esk 5mo ago
Yes, I work at a 50 person startup and even here switching from CC to codex or cursor would be non-trivial for multiple reasons - not just the annual commitment.
79.
by p1esk 5mo ago
What is “Sonnet 3.7 moment”?
80.
by p1esk 5mo ago
He deserves it. My guess is he demanded a piece of that 120M, didn’t get it, and leaked it to the press. My impression is police and FSB don’t get along too well in Russia, but both are equally corrupt.
81.
by p1esk 5mo ago
I’m curious why they didn’t arrest his family to get him, or to punish him. Maybe that was in exchange for him keeping his mouth shut?
82.
by p1esk 5mo ago
I didn’t understand that notation. Can someone please explain?
83.
by p1esk 5mo ago
I thought Grok is dead - its GPUs were handed over to Anthropic, no?
84.
by p1esk 5mo ago
I'm 1% in US.
85.
by p1esk 5mo ago
I’d look at software projects in automotive industry. Someone has to write software modern cars run, you can become that someone with enough focus and determination.
86.
by p1esk 5mo ago
This is my biggest concern. In the more distant future, I think people will lose themselves in VR worlds.
87.
by p1esk 5mo ago
The only big remaining problem in AI is continual learning. A lot of smart people are working on that. To me it looks like we are 1-2 breakthroughs away from AGI.
88.
by p1esk 5mo ago
It’s interesting, people talk about inequality and I definitely feel it myself – I see so many rich people around me. But I am in that 1%, just like many on this forum. At least according to https://dqydj.com/average-median-
89.
by p1esk 5mo ago
GPT1 through GPT3 advancement were exactly like using more Ethernet cables in parallel. All interesting conceptual breakthroughs came after GPT3: RL and reasoning being the main ones.
90.
by p1esk 5mo ago
Unless AI will allow people not work and keep their quality of life. Could be possible with total automation of everything.

More ›