Marcel Aach

I burn a lot of core hours to find out how to burn fewer core hours🔥 Large scale machine learning @FZJ_JSC👨🏻‍💻🇪🇺

[Thomas G. Dietterich @tdietterich]

If generative AI tools generate inappropriate language, plagiarized content, biased content, errors, mistakes, incorrect references, or misleading content, and that output is included in scientific works, it is the responsibility of the author(s). 2/

[Marcel Aach @ml_marce1]

@_onionesque Recently been seeing several job advertisements that mention the requirement to have published at Neurips/ICML/ICLR (main track) specifically.

[Marcel Aach @ml_marce1]

@giffmana Its insane, these people deliberately choose the „no LLM usage for reviews“ track and then still fell for some pretty simple prompt injections that were all over twitter and Reddit. Careless really is the right word here.

[Marcel Aach @ml_marce1]

@_xjdr literally every time I sit down to write the paper after running a zillion experiments I realize I missed some super obvious ablation that I never would've caught if I wasn't forced to organize my thoughts into coherent text for other humans to read

[Marcel Aach @ml_marce1]

@FSchaipp Very good and much needed work!

[Simo Ryu @cloneofsimo]

@fluxtheorist @ABAtanasov "You know the doomscroll business, and I know the distributed training at scale. Lets say, we partner up."

[Simo Ryu @cloneofsimo]

This is aging so well. We've reached the point where humans can make mind-drugs made out of bits that can (and will) hyper-optimize itself across time. Society will collapse if we poison our children. Stop this.

[Marcel Aach @ml_marce1]

@BlackHC I usually organize my stuff the same way John Conway did.

[Marcel Aach @ml_marce1]

@jxmnop „Scaling laws“ now mean essentially predictably evaluating on the test set.

[tender @tenderizzation]

when will AI engineers have an original idea

[Marcel Aach @ml_marce1]

@scaling01 Huh? That’s definitely not true, there are tons of research projects here that will use this machine for Foundation Model pre-training.

[Marcel Aach @ml_marce1]

@cloneofsimo Whats your take on Appendix D (basically saying dynamically varying the batch size is not really worth it?)

[Bojan Tunguz @tunguz]

I should start a nonprofit.

[Bojan Tunguz @tunguz]

Men would rather spend all of their life studying cellular automata than going to therapy.

[Tim Dettmers @Tim_Dettmers]

This model can automatically debug CUDA version errors. AGI achieved ✅😂

[Marcel Aach @ml_marce1]

@BlackHC "...dass der Essay ein außergewöhnliches Maß an Inhaltsdichte aufweise..."

Also das spricht doch eher gegen die Verwendung von ChatGPT. Auch fraglich wie sie auf die "zu 45 Prozent KI generiert" Wahrscheinlichkeit kommen.

[Marcel Aach @ml_marce1]

@StasBekman I wonder if there are more efficient ways of saving interesting things for later reading when you do not have time to do it the moment you find them.

[Marcel Aach @ml_marce1]

@owainkenway New machine incoming?

[Marcel Aach @ml_marce1]

@AndyHVandenBerg What’s the thing with Vail?

[Marcel Aach @ml_marce1]

@Rainmaker1973 We would have flying cars without this by now.