pama

17 ms·

271.
by pama 1y ago
Good performance improvement here: https://world-playground-deceit.net/blog/2025/03/speeding-up...
272.
by pama 1y ago
Although what you say about smoothing vs filtering is true in principle, for conditional generation of the eventual joint distribution starting from the same condition and using an autoregresive vs diffusive LLM, it is the smoothing distrib
273.
by pama 1y ago
I am still jetlagged and not sure what the most helpful reference would be. Maybe start from the block diffusion paper I recommended in a parallel thread and trace your way up/down from there. The logic leading to Eq 6 is a special ca
274.
by pama 1y ago
The human processing is still autoregressive, but using multiple parallel synchronized streams. There is no problem with such an approach and my best guess is that in the next year we will see many teams training models using such tricks fo
275.
by pama 1y ago
Less because pathogens multiply again once they are back in the clinic.
276.
by pama 1y ago
Was the name influenced by the ship in the murderbot diaries?
277.
by pama 1y ago
Unfortunately the intuition and the math proofs so far suggest that autoregressive training is learning the joint distribution of probabilistic streams of tokens much better than diffision models do or will ever do. My intuitive take is th
278.
Two Lessons from ICLR 2025(leon.bottou.org)
1 points by pama 1y ago|0 comments
279.
by pama 1y ago
Learning to formalize math to then prove Putnam competition problems rigorously in Lean would require you to have mid-to-advanced college level math and CS background. (Learning to do a small fraction of the Putnam competition without usin
280.
by pama 1y ago
People that belong in the job, even the creator of C++ himself, take decades to mature to the point of daring to use the language properly and then teach it. C++ is not meant to be learned in a week.
281.
ICLR meeting calendar (Singapore 2025)(iclr.cc)
2 points by pama 1y ago|0 comments
282.
by pama 1y ago
As a user of code I feel it is more permissive to me to see the code. As a licenser of code I want what is best for me and for others as a user and future developer. Others are free to do anything as long as they dont remove the license an
283.
by pama 1y ago
One could argue that GPL is very permissive. If you need to use it in a proprietary way in your own company for internal purposes, no problem; if you release software that others use, you have to release the code as well. I dont want to be
284.
by pama 2y ago
There is hope people will report this site and google will take it down quickly.
285.
by pama 2y ago
This smells like a quick attempt to enable phishing for vulnerabilities, and not a legit way to make progress. The comment is from a person that runs a security startup and the site is a google site that people can report to google as a sca
286.
by pama 2y ago
If its any consolation, having diagnosed a malignant papilary or follicular thyroid tumor (surprizingly) is correlated with equal or slightly increased lifespan compared to the population that did not have this tumor—maybe it is because of
287.
by pama 2y ago
Similar to how human brains behave, it is easier to train a model to select a better solution between many choices than to check an individual solution for correctness [1], which is in turn an easier task to learn than writing a correct sin
288.
by pama 2y ago
Hi Stavros. I would never say OMG—dont believe in deities—and I feel sorry that it came out that way. I was commenting on the GP, which was suggesting alternative ways to get similar effects to those of car-free cities, but in alternative
289.
by pama 2y ago
In modern cars, including EV, polution and noise come from weight and tires and breaks; the polution levels are a sharp funxtion of speed, but the speeds at which they dont matter much, dont offer benefits either. If we had magnetic/le
290.
by pama 2y ago
I have more than a dozen special commands for splitting to multiple n*m windows filled with different types of buffers or transposing existing window splits; I occasionally need a split into 4x4 or 5x5 or even more windows to quickly/v
291.
by pama 2y ago
Thanks.
292.
by pama 2y ago
Thanks!
293.
by pama 2y ago
Wouldn’t fixing this bug reduce security?
294.
by pama 2y ago
Not sure what you mean about ubuntu… there is tons of GPL there. https://ubuntu.com/legal/open-source-licences?release=jammy
295.
by pama 2y ago
No it didn’t. At least not for new small molecule drugs. It did reduce times a bit for the first vaccines because there were many volunteers available, and it did allow some antibody drug candidates to be used before full testing was comple
296.
New Windows Command for Emacs31(p.bauherren.ovh)
3 points by pama 2y ago|0 comments
297.
by pama 2y ago
“But if AI vanished tomorrow due to, say, regulations […]” You can’t vanish math and computers globally like that.
298.
by pama 2y ago
Embrace the confusion with the AI—it’s a sign of progress!
299.
by pama 2y ago
Optuna often works fine in this context (even with the memory errors or, with some tuning, with the non-halting runs): https://github.com/optuna/optuna
300.
Supercharge DeepSeek-R1 Inference on AMD Instinct MI300X(rocm.blogs.amd.com)
2 points by pama 2y ago|0 comments

More ›