ipsum2

49 ms·

751.
by ipsum2 4y ago
Cerebras makes impressive hardware, but Nvidia still performs better in every regard. The most telling factor is that Cerebras claims they're too busy to run common benchmarking (e.g. MLPerf) to compare against Nvidia.
752.
by ipsum2 4y ago
He recently gave a talk that covers robotics for a general audience: https://www.youtube.com/watch?v=f6fDpKDxpL0
753.
by ipsum2 4y ago
Owned by HPE: https://www.hpe.com/us/en/newsroom/press-release/2021/06/hew... Looking through the documentation, the API looks brittle. I'll stick with slurm for large jobs and run things
754.
by ipsum2 4y ago
Machine learning is so hyped right now (with good reason) so customers are price insensitive.
755.
by ipsum2 4y ago
I would sell a kidney for one of these. It's basically impossible to train language models on a consumer 24GB card. The jump up is the A6000 ADA, at 48GB for $8,000. This one will probably be priced somewhere in the $100k+ range.
756.
by ipsum2 4y ago
The CEO and founder of Twitch announced a week ago that he was leaving the company. It's admirable to see a captain going down with his ship. https://news.ycombinator.com/item?id=35184011
757.
by ipsum2 4y ago
1) It is a general version of knowledge distillation. For example, this paper from 2016 describes the same technique: Sequence-Level Knowledge Distillation [0] > This sequence-level approximation leads to a simple training procedure wher
758.
by ipsum2 4y ago
I thought the implications were pretty obvious. OpenAI prohibits this in their TOS for explicitly this purpose and is called out in the Alpaca post[0] > You may not ... use the Services to develop foundation models or other large scale m
759.
by ipsum2 4y ago
Yudkowsky has a very shallow understanding of machine learning/AI, but loves to write in a conspiratorial way with in-group terminology to make people think that he's onto some dark secrets. What he's describing is common kno
760.
by ipsum2 4y ago
It is either intentional or awfully coincidental that that Google would announce this two days before Microsoft's AI event: https://news.microsoft.com/reinventing-productivity/ , which will probably discuss similar
761.
by ipsum2 4y ago
Is this currently used at Google?
762.
by ipsum2 4y ago
How did you deal with data contamination?
763.
by ipsum2 4y ago
> run on a custom saliva-resistant touch-screen console So a standard resistive touch screen?
764.
by ipsum2 4y ago
No they haven't. Stability AI is funded by the founder and VC money, not crowdsourcing.
765.
by ipsum2 4y ago
Have you tried training a large model before? If not, you're probably discounting how difficult and expensive it is.
766.
by ipsum2 4y ago
Can you discuss the technical details of how this was implemented? From a high level perspective, I can imagine it involves creating a 3d LUT from a color chart.
767.
by ipsum2 4y ago
Most media outlets downplayed COVID in early 2020. Now they've over-corrected and report on every minor outbreak and stoking fears for views and ad revenue.
768.
by ipsum2 4y ago
AGI is usually defined as anything a normal human could do. Examples of some things that GPT is currently lacking: 1) Understanding images and video 2) Learning and remembering things outside the 2048 token context window 3) Interacting wit
769.
by ipsum2 4y ago
That's an inaccurate description of the paper. The "shocks" are continuous, not reward driven. Both control and experimental groups received the electrical current.
770.
by ipsum2 4y ago
Unless you're wearing an EEG device, it's probably different from what's being studied here. Audio entrainment without feedback has been known for 40+ years and hasn't shown meaningful benefits AFAIK.
771.
by ipsum2 4y ago
That's a ridiculous forecast.
772.
by ipsum2 4y ago
There are 11 startups named "Artifact", and lots more with Artifact in their name (e.g. Artifact Technologies) https://www.crunchbase.com/search
773.
by ipsum2 4y ago
Is the full sized image posted anywhere? This could just be the limits of dynamic range blowing out the text in the iPhone pics.
774.
by ipsum2 4y ago
FYI to people unaware, the author of this piece, Chamath is an opportunist and possibly commits fraud. https://www.newyorker.com/magazine/2021/06/07/the-pied-piper... https://www.protocol.com&
775.
by ipsum2 4y ago
Getting reliable GPUs is a difficult problem, I empathize. I've spent a decent amount of time and money because there was one failing GPU on an AWS cluster.
776.
by ipsum2 4y ago
Note that this doesn't take into account the numerous iterations required to dial in the correct hyperparameters and model architecture, which could easily increase cost 5-10x. > 256 A100 throughput was extrapolated using the other
777.
by ipsum2 4y ago
Noticed Github and Imgur acting up and came to HN to figure out why. Thanks for the update.
778.
by ipsum2 4y ago
It's pretty clear from the article that 8-bit inference is referring to FP8, as the author claims int8 is similar to fp16 inference performance. > Ada/Hopper also have FP8 support, which makes in particular 8-bit training much
779.
by ipsum2 4y ago
The graphs ranking GPUs may not be accurate, as they don't represent real-world results and have factual inaccuracies. For example: > Shown is raw relative performance of GPUs. For example, an RTX 4090 has about 0.33x performance of
780.
by ipsum2 4y ago
I didn't read it as hateful, but as legitimate curiosity. Ruby is at a disadvantage because the analytics ecosystem is primarily Python based.

More ›