apsec112

24 ms·

331.
by apsec112 3y ago
That... doesn't really explain how they can get such a high number? Standard FLOP efficiency on fine-tuning big models is like 30-40%. How can you get 750%?
332.
by apsec112 3y ago
I haven't run the code, but... how is this even possible? I've done PyTorch profiling on QLoRA Llama-2-70B fine tunes, and the runtime is dominated by the large matrix multiplies in the MLP layers, plus a bit for attention. Under
333.
The Paperwork Reduction Act Is Terrible(goodscience.substack.com)
3 points by apsec112 3y ago|0 comments
334.
Ilya Sutskever "at the center" of Altman firing?(twitter.com)
416 points by apsec112 3y ago|488 comments
335.
More Dakka in Medicine (2019)(qualiacomputing.com)
54 points by apsec112 3y ago|28 comments
336.
Bangladesh saved lives by getting lead out of turmeric spice(vox.com)
133 points by apsec112 3y ago|119 comments
337.
Casino-like apps have drained people of millions(nbcnews.com)
291 points by apsec112 3y ago|396 comments
338.
by apsec112 3y ago
I use L40s because H100s and A100s are backlogged, but for big models the lack of memory capacity and PCIe card-communication bottleneck is a pain, I'm planning to switch as soon as there's capacity available
339.
MetNet-3: A state-of-the-art neural weather model(blog.research.google)
180 points by apsec112 3y ago|72 comments
340.
Pretraining on the Test Set Is All You Need(arxiv.org)
68 points by apsec112 3y ago|26 comments
341.
U.S. GDP grew at a 4.9% annual pace in the third quarter, better than expected(cnbc.com)
283 points by apsec112 3y ago|697 comments
342.
The Moose Boulder Hoax(atlasobscura.com)
55 points by apsec112 3y ago|14 comments
343.
by apsec112 3y ago
That was California only, this is federal
344.
Trade Regulation Rule on Unfair or Deceptive Fees [pdf](ftc.gov)
2 points by apsec112 3y ago|0 comments
345.
When Product Markets Become Collective Traps: The Case of Social Media(bfi.uchicago.edu)
1 points by apsec112 3y ago|0 comments
346.
by apsec112 3y ago
* as a percentage of US GDP, not in terms of actual dollars
347.
Llama 2 Long(arxiv.org)
134 points by apsec112 3y ago|31 comments
348.
by apsec112 3y ago
"They are the investors of large proprietary AI companies" is just... not true? Not sure where you're even getting this from. I'm a modestly successful upper-middle-class ML engineer, and I've been worried about AI
349.
by apsec112 3y ago
This comment describes the research: https://news.ycombinator.com/item?id=37635401
350.
by apsec112 3y ago
Almost everyone is afraid of death. Like, if you point a gun at someone, most people will feel fear. The main exceptions are people who are so sick with age-related diseases that they can't live happy lives anymore, but, of course, agi
351.
by apsec112 3y ago
Thanks! They have a deal with Hillman for use of the tech, it's in the slides: https://docs.google.com/presentation/d/1OqDqFYMQdcS0XM4hORqN...
352.
by apsec112 3y ago
The average time from scientific discovery to commercial success in biotech is measured in decades. Barry Werth's book The Antidote goes into why it's so slow and complicated: https://www.amazon.com/Antidote-Inside
353.
by apsec112 3y ago
Longevity biotech came from a similar milieu and has been doing very well: https://english.elpais.com/economy-and-business/2023-07-17/t... (along with AI companies of course)
354.
by apsec112 3y ago
Hi! These are friends of mine. There is much more detail in the Google Drive (linked from the About page): https://drive.google.com/drive/u/2/folders/18ZDSe92LgLmS0sUb...
355.
by apsec112 3y ago
The amounts produced are tiny. See the safety FAQ here: https://docs.google.com/document/d/1mDJCTO2QySmQOZQcajYReDCA...
356.
by apsec112 3y ago
The amounts produced are tiny. See the safety FAQ here: https://docs.google.com/document/d/1mDJCTO2QySmQOZQcajYReDCA...
357.
Gene-Engineered Mouth Bacteria(lanternbioworks.com)
416 points by apsec112 3y ago|296 comments
358.
by apsec112 3y ago
It's not, it's just a re-hosted clone of the original Enron site circa 2000 ( https://web.archive.org/web/20011003192049/http://www.enron.... ).
359.
by apsec112 3y ago
Testimony is often wrong. The testimony for the Mormon religion is just as compelling (arguably more compelling) than any UFO incident, yet the vast majority of people don't convert to Mormonism. https://archive.ph/RsSt
360.
by apsec112 3y ago
The vast majority of greenhouse gas emissions isn't military, it's for ordinary consumer uses like heating, cooling, driving, and farming meat. No matter what the form of government, lots of people will object very strongly if the

More ›