nialv7

20 ms·

301.
by nialv7 2y ago
OpenAI has an incentive to make people believe that the scaling laws are still alive, to justify their enormous capex if nothing else. I wouldn't give what they say to much credence, and will only believe the results I see.
302.
by nialv7 2y ago
Launching from that high of a latitude feels like it could be a problem. Basically only the polar orbits will be available.
303.
by nialv7 2y ago
Oh my, so much cope and denial here...
304.
by nialv7 2y ago
You need to be prepared for the reality that naive scaling no longer works for LLMs anymore. Simple question: where is GPT-5?
305.
by nialv7 2y ago
Yes and no. In their paper they said they trained two models. One is purely RL based (R1Zero). So this one is trained like you described, i.e. it has to stumble upon the correct answer. They found it to be good but has problems like repetit
306.
by nialv7 2y ago
I think the natural-virus-leaked-by-lab theory hinges on the argument that, (assuming it was true, then) had the lab leak not have happened, SARS-CoV-2 wouldn't have made to jump to human by itself. And this is where the Ebola analogy
307.
by nialv7 2y ago
I was curious so I did a quick research on the previous SARS-CoV-1, the one that caused an outbreak back in 2003. Looks like they weren't able to find the natural reservoir for that one either. We did know it came directly from masked
308.
by nialv7 2y ago
Very good summary, but the problem is, like most things, it is impossible to 100% rule out the lab theory. It is always possible to look at one of these evidences and say "yeah, but ...". And given how politicized this question is
309.
by nialv7 2y ago
o1-preview was released Sep 12, 2024. So DeepSeek team probably had a couple of months.
310.
by nialv7 2y ago
I'd bet $5 that o1 was also built with either RL or search, or a combination of the two. That was what I initially thought when they announced o1-preview, after I saw the sample reasoning traces. But alas I am just an ML enthusiast, no
311.
by nialv7 2y ago
The group in this article used straight and simple PPO, so I guess GRPO isn't required. My hypothesis is that everyone was just so stunned by oai's result so most just decided to blindly chase it and do what oai did (i.e. scaling
312.
by nialv7 2y ago
Lock files _are_ code. Same for meson.build or CMakeList.txt. We don't use to have the ability to precisely specify dependencies in code, now we do. Building with a patched codebase is effectively building a fork.
313.
by nialv7 2y ago
If a distro chose to build some program with a different set of dependencies than what is specified by its author, then arguably it is not the same piece of software anymore. If Debian wants to do that - which I think they 100% have the rig
314.
by nialv7 2y ago
I am quite tired of seeing titles like this. No, you _don't know_. Vast majority of definitive statements like this is going to be meaningless. The whole point about this is that it's an uncertainty, the impact of AI on our societ
315.
by nialv7 2y ago
This made me lose a bit of faith in humanity, because it seems like when a group of humans get together, no matter what kind of structure the group take, it's going to be bad in some way. Hierarchical, flat, or structureless (as pointe
316.
by nialv7 2y ago
that's not how this works. o1's thinking trace is hidden, and that's what's valuable here, not the output.
317.
by nialv7 2y ago
you can't get o1's thinking trace I believe?
318.
by nialv7 2y ago
I paid ~$750 for my 350mm Voron 2.4 kit (and, sure, 40 hours of my time. But look, you want to do 3D printing, 40 hours are just a small initial investment).
319.
by nialv7 2y ago
bait-and-switch? We, those who advocate for open source 3D printers, saw it coming from miles away. This has very very clearly been their plan all along, they themselves said as much (e.g. they are doing the "apple model"). They h
320.
by nialv7 2y ago
Wouldn't it generate more heat than is needed to heat the fuel mixture?
321.
by nialv7 2y ago
Never expected to see the EFF siding with a big tech company, and fighting for its right to profit from its users. Never expected to see the EFF dismiss an argument for user's data privacy as "shaky". Quite disappointed hones
322.
by nialv7 2y ago
Two questions I have: 1. How much of the fuel's energy is released as heat? They have a heat recapture device, but that's only used to preheat air/fuel, and not used to generate electricity. Is the energy in the heat simply d
323.
by nialv7 2y ago
do we know when this video was taken? this could just be ship breaking up during re-entry because it lost altitude control. not necessarily the moment of the primary failure.
324.
by nialv7 2y ago
You said what I wanted to say. I think there is a niche market for tools that allow individuals to train their own recommendation systems.
325.
by nialv7 2y ago
So, let me get this straight. Scraping data and publishing it so everyone get equal and free access to information and knowledge, arguably making the world better, is a crime. But scraping it to benefit and enrich just yourself, is A-OK?? T
326.
by nialv7 2y ago
oh, that's interesting. do UNC paths not have to be valid UTF-16?
327.
by nialv7 2y ago
Windows really should provide an API that treats path names as just bytes, without any of these stupid encoding stuff. Could probably have done that when they introduced UNC paths.
328.
by nialv7 2y ago
one thing you can reason about a function is: does it exist at all? if you don't have parametricity, you can't even be sure about that. in Rust, as long as your type satisfies a generic function's bounds, you can be sure inst
329.
by nialv7 2y ago
bad analogy. a better one would be: a wheel fell of an F1 car, car hit perimeter wall, driver dies. should we maybe put a crash barrier in front of the wall?
330.
by nialv7 2y ago
yeah, this is so simple, i don't understand why people keep invent more complex and worse rules to understand cdecl...

More ›