Stefan Bachhofner @fcbstef

Joined December 2011

Tweets

2K
Followers

146
Following

4K
Likes

7K

Dimitris Papailiopoulos @DimitrisPapail

3 weeks ago

Thinking about model generalization is quite painful. We observe empirically that models trained with SGD on cross-entropy generalize, instead of just memorize the train data. Even when they have sufficient capacity to memorize. We do not---i repeat--- we. do. not. have a…

74 57 728 78K 523

Emanuele Marconato @ema_marconato

3 months ago

🧵Why are linear properties so ubiquitous in LLM representations? We explore this question through the lens of 𝗶𝗱𝗲𝗻𝘁𝗶𝗳𝗶𝗮𝗯𝗶𝗹𝗶𝘁𝘆: “All or None: Identifiable Linear Properties of Next-token Predictors in Language Modeling” Published at #AISTATS2025🌴 1/9

8 61 325 29K 303

Download Image

Heike Lehner @heikelehner

3 months ago

Ein großes Problem für Österreichs Produktivität und Wachstum ist auch die unterdurchschnittliche Unternehmensdynamik. Der demografische Wandel machts nur noch schlimmer. produktivitaetsrat.at

3 2 7 641 2

Download Image

Andreas Kirsch 🇺🇦 @BlackHC

3 months ago

I'm late to review the "Illusion of Thinking" paper, so let me collect some of the best threads by and critical takes by @scaling01 in one place and sprinkle some of my own thoughts in as well. The paper is rather critical of reasoning LLMs (LRMs): x.com/MFarajtabar/st…

Mehrdad Farajtabar @MFarajtabar

3 months ago

111 582 3K 879K 4K

Download Image

33 227 1K 432K 2K

Ricky Gervais @rickygervais

3 months ago

I did this cameo for a laugh when I was visiting the set. It wasn’t in the script & there was no rehearsal. We just filmed 2 idiots ad libbing. What a fun morning.

720 9K 192K 8.5M 22K

Download Video

Jeremy Howard @jeremyphoward

3 months ago

When I was optimising ULMFiT, I came up with a trick where I ran lots of ablations and fed all the hyperparams and results to a random forest. That told me which were most important. I told @l2k about it, and @weights_biases added it to their product! :D youtu.be/2QX6jjMt1Eg?si…

6 15 246 18K 139

Download Image

MIT CSAIL @MIT_CSAIL

3 months ago

GenAI is almost 5x less accurate than humans when summarizing scientific research, according to new study: bit.ly/4dzcDMT

22 159 646 46K 286

Download Image

Leonardo Rizzo @LnrdRizzo

4 months ago

How we "guessed" the Pope using network science: inside the cardinal network. A study by me, Beppe Soda and Alessandro Iorio. Article: unibocconi.it/en/news/networ… @Unibocconi

237 2K 12K 2.0M 8K

Download Image

LaurieWired @lauriewired

3 months ago

Want to recognize a song from just a few seconds of distorted audio? Use Constellation Maps. The math is brilliantly simple. With just a handful of bytes; discarding 99% of the waveform, you can recognize a unique fingerprint across hundreds of millions of tracks.

41 355 4K 167K 3K

Download Image

Friedemann Zenke @hisspikeness

3 months ago

1/6 Why does the brain maintain such precise excitatory-inhibitory balance? Our new preprint explores a provocative idea: Small, targeted deviations from this balance may serve a purpose: to encode local error signals for learning. biorxiv.org/content/10.110… led by @j_rossbroich

6 39 247 16K 182

Download Image

Tim Dettmers @Tim_Dettmers

3 months ago

MXFP4 and MXFP8 matmuls will be the standard for training models for the foreseeable future -- this work provides the first full implementation. Scaling looks much better than regular FP datatypes, but the next generation of agent models is likely still optimal in 8-bit.

Dan Alistarh @DAlistarh

3 months ago

20 78 397 62K 241

Download Image

4 27 188 18K 91

Taco Cohen @TacoCohen

3 months ago

Nobody wants to hear it, but working on data is more impactful than working on methods or architectures.

Katie Everett @_katieeverett

4 months ago

Nobody wants to hear it, but working on data is more impactful than working on methods or architectures.

16 93 627 243K 603

30 96 977 160K 306

Aidan McLaughlin @aidan_mclau

4 months ago

god i love capitalism

379 2K 21K 3.3M 6K

Download Video

Heike Lehner @heikelehner

4 months ago

Extrem wichtiges Thema am Cover vom aktuellen Economist. "What should Europeans do? They should start by returning to the old liberal ideas that *noisy disagreement is better than enforced silence* and that people should tolerate one another’s views." economist.com/leaders/2025/0…

8 31 102 16K 11

Download Image

Chris Freiman @cafreiman

4 months ago

A reminder that defenders of capitalism don’t actually accept: Trickle-down economics That the wealthy are entitled to their wealth because they’re morally better That hard work alone is sufficient for success That we’re atomistic individuals with no need for community

59 103 737 22K 52

Peyman Milanfar @docmilanfar

4 months ago

Mean-shift iteratively moves points towards regions of higher density. It does so by placing a kernel at each data point, calculating the mean of the data points within that window, shifting points towards this mean until convergence: Look familiar? 1/n (anim by @gabrielpeyre)

13 72 556 47K 426

Download Video

Heike Lehner @heikelehner

4 months ago

Bei 45,5 % Abgabenquote "Optimismus und Zuversicht" zu verlangen, ist schon ein bisschen #delulu, wie die coolen Kids sagen würden.

3 7 75 4K 0

Heike Lehner @heikelehner

4 months ago

Solange in Österreich die heiligen Kühe, ein wenig nachhaltiges Pensionssystem, überbordende Bürokratie und zu hohe Arbeitskosten, nur gestreichelt statt geschlachtet werden, kommen wir aus der Misere nicht heraus.

4 12 96 3K 0

Heike Lehner @heikelehner

4 months ago

#budgetrede was Österreich bzw der gesamten EU wirklich helfen würde, ist, wenn die massiven EU-internen Handelsbarrieren abgebaut werden würden - für Güter sind die äquivalent zu einem 45%-Zoll, für Dienstleistungen 110%.