Gal Katzh (@GalKatzh) header
Gal Katzh (@GalKatzh) profile picture

the adjacent possible | https://teapaw.org/t.co/PpnMYVmZ29 | @gk@sigmoid.social

Israel
Joined July 2008
Photo by Gal Katzh (@GalKatzh)Photo by Gal Katzh (@GalKatzh)Photo by Gal Katzh (@GalKatzh)Photo by Gal Katzh (@GalKatzh)Photo by Gal Katzh (@GalKatzh)Photo by Gal Katzh (@GalKatzh)Photo by Gal Katzh (@GalKatzh)Photo by Gal Katzh (@GalKatzh)
Iโ€™m joining the Weizmann Institute as an Assistant Professor (known[?] as โ€œSenior Scientistโ€ ๐Ÿง‘โ€๐Ÿ”ฌ) and starting CoLab๐Ÿถ! We do pretraining, agents and evaluation the way science should be done: in a community, together. Come research with us ๐Ÿ‘‡ discord.gg/gVeGEjWTtv
37
20
2
393
11,479
ื ืชื ื™ื”ื• ื›ืขืช ื‘ืงืžืคื™ื™ืŸ "ื”ืขืจื‘ื™ื ื’ื ื‘ื• ืืช ื”ื‘ื—ื™ืจื•ืช". ื‘ืžื˜ืจื” ืœื™ื™ืฆืจ ื“ื” ืœื’ื™ื˜ื™ืžืฆื™ื” ืœื‘ื—ื™ืจื•ืช, ืขื•ืจื‘ื• ื‘ืขื ื™ื™ืŸ ืจืืฉ ืฉื‘"ื› ื–ื™ื ื™, ืžืคื›"ืœ ื”ืžืฉื˜ืจื” ื“ื ื™ ืœื•ื™ ื•ื™ื•"ืจ ื•ืขื“ืช ื”ื‘ื—ื™ืจื•ืช ืกื•ืœื‘ืจื’. ืืช ื”ืจืื™ื•ืช ืœื–ื™ื•ืฃ ื›ื‘ื™ื›ื•ืœ ืฉืœ ื”ื‘ื—ื™ืจื•ืช, ืžืกืจ ืฉื’ื™ื ืงื™ื™ื–ืœืจ โ€“ "ืžื ื”ืœ ื˜ื•ื”ืจ ื”ื‘ื—ื™ืจื•ืช ื‘ืžื’ื–ืจ ื”ืขืจื‘ื™ ืžื˜ืขื ื”ืœื™ื›ื•ื“". ืžื™ ื”ืื™ืฉ ืขื ื”ืชื•ืืจ ื”ื—ืฉื•ื‘? ืงื™ื™ื–ืœืจ ื”ื•ื ืœื ืื™ืฉ ื—ื•ืง ืื• ืฉื•ืคื˜ ื‘ื“ื™ืžื•ืก. ืžื“ื•ื‘ืจ ื‘ืงืžืคื™ื™ื ืจ ืฉืžืชืžื—ื” ื‘ืคืขื™ืœื•ืช ืœื“ื™ื›ื•ื™ ื•ื“ื” ืœื’ื™ื˜ื™ืžืฆื™ื” ืœื‘ื—ื™ืจื•ืช ื‘ื—ื‘ืจื” ื”ืขืจื‘ื™ืช. ื‘ืฆื™ืœื•ื ื”ืžืกืš: ืžืฉืจื“ ื”ืงืžืคื™ื™ื ื™ื "ืงื™ื™ื–ืœืจ ืขื ื‘ืจ" ืžืชื’ืื” ื‘ื”ืฆืœื—ืชื• ืœื”ื‘ืขื™ืจ ืืช ื”ืฉื˜ื— ื‘ืงืžืคื™ื™ืŸ ื”ื‘ื—ื™ืจื•ืช ื‘-2019
47
333
43
1,320
69,446
Gal Katzh retweeted
1/ What is the maximum number of legal moves in a chess position where every one of them gives check? The record on that question had stood since 1968. Double Agent, our AI research system, just broke it. How many checks can you count here?
2
12
7
38
3,360
Gal Katzh retweeted
ื”ืชื ืคืœื•ืช ืฉื•ื“ื“ื™ื ื‘ืืจืฅ ื™ืฉืจืืœ ื ื“ื™ืจื” ืžืื•ื“. ืžืจื‘ื™ืช ื”ืกื™ืคื•ืจื™ื ืขืœ ื›ืš ืขืœ ื™ื“ื™ ืชื™ื™ืจื™ื-ื ื•ืกืขื™ื ืื™ื ื ืžื‘ื•ืกืกื™ื ืืœื ืขืœ ืื”ื‘ืช ื”ืจื•ืžื ื˜ื™ืงื” ืœืฉืžื”, ื•ืชื•ืฆืืชื ื”ื™ื—ื™ื“ื” ืžืฆื˜ืžืฆืžืช ื‘ื‘ืงืฉื™ืฉ ื ื›ื‘ื“ ื”ืžืฉื•ืœืฉืœ ืœืื ืฉื™ ื”ืžืฉืžืจ [ืœืื—ืจ ื‘ื™ื•ื ืฉื•ื“ ื›ื–ื” ืขื‘ื•ืจ ื”ืชื™ื™ืจื™ื].
2
1
16
381
ืกื•ืฃ ื›ืœ ืกื•ืฃ
Today @GoogleMaps is getting its biggest upgrade in over a decade. By combining our Gemini models with a deep understanding of the world, Maps now unlocks entirely new possibilities for how you navigate and explore. Hereโ€™s what you need to know ๐Ÿงต
98
Open syllabus ๐Ÿค Annaโ€™s archive
1
2
618
Inspiring recap for the LLM jailbreaking competition (winning defense team!)
My team with @Yuvlem won๐Ÿฅ‡first place in the defense track of the Capture The Flag-LLM Challenge at @satml_conf. In the competition, each team hides a secret in an LLM context, and other teams try to discover it. Here's how we used ideas from Judo and Cooking to do it๐Ÿ‘‡ ๐Ÿงต1/11
1
219
Fascinating work by the brilliant @eladgranot and my PI, illuminating the importance of element-wise activation functions, using a powerful, little known technique for yielding the upper bound
Thrilled to share our paper with Amit Daniely at the ALT 2024 conference! We prove a tight generalization bound for two-layer networks, shedding light on the role of the element-wise property of activation functions in achieving these bounds. โœจ #ALT2024 arxiv.org/abs/2211.09634
3
204
Gal Katzh retweeted
Can we design good transaction fee mechanisms (TFMs)? If so, what are the trade-offs? Together with @yaish_aviv, we show an impossibility for deterministic TFMs, and limitations for randomized TFMs. Paper: doi.org/10.48550/arXiv.2402.โ€ฆ Read on for more details (1/10)
4
21
3
74
10,005
Wow! This might be the most interesting detail about Palestinian conflict deaths: they're basically just male. Israel must not be indiscriminate.
Replying to @cremieuxrecueil
There's also data on the Israeli age distribution (they don't specify whether or not victims took part in hostilities), so I split it up by gender. I'll include the Palestinian one split by Gender as well for comparison.
46
162
24
1,076
256,363
Did you think your model is protected if you publish it as a black box? Think twice. Happy to share our #ICLR2023 paper: An Exact Poly-Time Membership-Queries Algorithm for Extraction a three-Layer ReLU Network arxiv.org/abs/2105.09673 w. Amit Daniely
15
27
1,537
The sound of machine learning that is just logistic regression
57
1,817
298
5,426
We want to pretrain๐Ÿคž Instead we finetune๐Ÿšฎ๐Ÿ˜” Could we collaborate?๐Ÿค— ColD Fusion: ๐Ÿ”„Recycle finetuning to multitask โžก๏ธevolve pretrained models forever On 35 datasets +2% improvement over RoBERTa +7% in few shot settings ๐Ÿงต #NLProc #MachinLearning #NLP #ML #modelRecyclying
6
25
13
128
This meme implies weโ€™re living in uniquely stupid times, this 1994 letter to an editor begs to differ:
2
8
1
35
Proud to be a data point in this cool research
Analogies are everywhere! We Gotta Catch 'Em All! โ—“๐Ÿ˜Ž ๐Ÿ“ฃ #emnlp2022 (main conference) long paper ๐ŸŽ‰: ๐ŸŽช Life is a Circus and We are the Clowns: Automatically Finding Analogies between Situations and Processes ๐Ÿคก w/ @HyadataLab paper: arxiv.org/abs/2210.12197 ๐Ÿงต 1/
1
2
Transformers are not Seq2Seq Given a context They predict a single token We use this to update representations between predicted tokens & feed a changing graph relevant to the current token arxiv.org/abs/2101.12640 me @oabend #conll2020 #NLProc
2
19
1
94
A few hours ago I had a discussion with friends about the state of ethics in CS academy. I was on the defending side. Seeing a leading prof dismissing such a burning ethical issue (which is a gray area imho), I realize I might have been completely wrong.
Replying to @Grady_Booch
"I don't want my Image to be indexed." Then. Don't. Post. Your. Image. (Or block the bot.)
1
1
2
Oof, the irony. The masterful โ€œAI artistโ€ prompt-writers donโ€™t want a machine or anyone else learning their โ€œcraftโ€ ๐Ÿ™ƒ.
127
3,684
302
20,577