Matt (@matthieubulte) 's Twitter Profile
Matt

@matthieubulte

math ∪ stats ∪ programming

ID: 2326897792

calendar_today05-02-2014 22:40:50

1,1K Tweet

1,1K Followers

568 Following

Matt (@matthieubulte) 's Twitter Profile Photo

Trying to optimize runtime of some simulation for my latest paper - numpy: 2.7s/it - numba: 7.5it/s * - jax: 109it/s 💪 * without parallelization, I couldn't get rid of the bugs in the code generated by parallel numba 🤷‍♂️

Nabeel S. Qureshi (@nabeelqu) 's Twitter Profile Photo

The prompt in question literally tells o1 "nothing else matters" and to achieve its goal at all costs. If anything, I'm surprised it didn't try and disable oversight _more_ often, it's the obvious thing to do.

The prompt in question literally tells o1 "nothing else matters" and to achieve its goal at all costs. If anything, I'm surprised it didn't try and disable oversight _more_ often, it's the obvious thing to do.
Matt (@matthieubulte) 's Twitter Profile Photo

Great piece from Patrick McKenzie on debanking - actually explains what's going on instead of the hysteria we've witnessed around here bitsaboutmoney.com/archive/debank…

Luigi Gresele (@luigigres) 's Twitter Profile Photo

This was fantastic news. Thanks ELLIS! My thesis, "Learning Identifiable Representations: Independent Influences and Multiple Views", can be found here: tobias-lib.ub.uni-tuebingen.de/xmlui/handle/1…

Johannes Hingerl (@thisisjohahi) 's Twitter Profile Photo

Introducing Flashzoi⚡! We’ve upgraded the Borzoi model with rotary pos. encodings and FlashAttention, resulting in a 3x speedup with similar or better accuracy for faster variant effect prediction or model development, and more efficient genomic analysis biorxiv.org/content/10.110…

Introducing Flashzoi⚡! We’ve upgraded the Borzoi model with rotary pos. encodings and FlashAttention, resulting in a 3x speedup with similar or better accuracy for faster variant effect prediction or model development, and more efficient genomic analysis biorxiv.org/content/10.110…
Corry Wang (@corry_wang) 's Twitter Profile Photo

Not something I see a lot of people talk about - but nuclear fusion might be the best historical example of a technology where a scaling law ultimately failed Fusion reactors were improving by 10x every 5 years until the mid-80s… and then basically flatlined into the 2000s

Not something I see a lot of people talk about - but nuclear fusion might be the best historical example of a technology where a scaling law ultimately failed

Fusion reactors were improving by 10x every 5 years until the mid-80s… and then basically flatlined into the 2000s
Matt (@matthieubulte) 's Twitter Profile Photo

Cute little lemma that I proved and added to the intro of my thesis. It's almost trivial, but now completely, and I couldn't find a proof elsewhere.

Cute little lemma that I proved and added to the intro of my thesis. It's almost trivial, but now completely, and I couldn't find a proof elsewhere.
Matt (@matthieubulte) 's Twitter Profile Photo

From the arm neon doc, it doesn't even say which register is being multiplied/added, you need to dig in the pseudo-code on some other page. Is it because there is a standard in the order of the arguments with dst always being the first?

From the arm neon doc, it doesn't even say which register is being multiplied/added, you need to dig in the pseudo-code on some other page. Is it because there is a standard in the order of the arguments with dst always being the first?
Matt (@matthieubulte) 's Twitter Profile Photo

Close enough to being done with my thesis that I now feel comfortable running all the updates that have accumulated over the last months.

Matt (@matthieubulte) 's Twitter Profile Photo

Last late evening hack: building a Jupyter magic for C programming. It persists state between cells, giving the full interactive notebook experience for low-level programming.

Last late evening hack: building a Jupyter magic for C programming. It persists state between cells, giving the full interactive notebook experience for low-level programming.