Yuval Rabani retweeted
OpenAI has launched Operator, an agent that can perform tasks in your browser. I asked it to complete a Qualtrics survey I created. The results are very promising for Operator but *very* concerning for survey researchers
Yuval Rabani retweeted
The presidents of @Harvard, @MIT, and @Penn were all asked the following question under oath at today’s congressional hearing on antisemitism:
Does calling for the genocide of Jews violate [your university’s] code of conduct or rules regarding bullying or harassment?
The answers they gave reflect the profound moral bankruptcy of Presidents Gay, Magill and Kornbluth.
Representative @EliseStefanik was so shocked with the answers that she asked each of them the same question over and over again, and they gave the same answers over and over again.
In short, they said:
It ‘depends on the context’ and ‘whether the speech turns into conduct,’ that is, actually killing Jews.
This could be the most extraordinary testimony ever elicited in the Congress, certainly on the topic of genocide, which to remind us all is:
“the deliberate killing of a large number of people from a particular nation or ethnic group with the aim of destroying that nation or group”
The presidents’ answers reflect the profound educational, moral and ethical failures that pervade certain of our elite educational institutions due in large part to their failed leadership.
Don’t take my word for it.
You must watch the following three minutes. By the end, you will be where I am.
They must all resign in disgrace.
If a CEO of one of our companies gave a similar answer, he or she would be toast within the hour.
Why has antisemitism exploded on campus and around the world?
Because of leaders like Presidents Gay, Magill and Kornbluth who believe genocide depends on the context.
To think that these are the leaders of Ivy League institutions that are charged with the responsibility to educate our best and brightest.
On the bright side, our congressional leaders deserve accolades for showing tremendous leadership and moral clarity in their statements, by the questions they asked, and the respectfulness with which they conducted the hearing.
It was a masterclass of how our government and democracy should operate.
If you have time, please watch the entire hearing. Throughout the hearing, the three behaved like hostile witnesses, exhibiting a profound disdain for the Congress with their smiles and smirks, and their outright refusal to answer basic questions with a yes or no answer.
Yuval Rabani retweeted
Less than 31 hours since OpenAI started dropping the ChatGPT vision feature on pro users...
People are scratching their heads in disbelief.
10 wild examples:
Yuval Rabani retweeted
Last couple of weeks I gave a few talks on the Sparks paper, here is the MIT recording!
The talk doesn't do justice to all the insights we have in the paper itself. Neither talk nor twitter threads are a substitute for actual reading of the 155 pages :-)
youtube.com/watch?v=qbIk7-JP…
Yuval Rabani retweeted
At @MSFTResearch we had early access to the marvelous #GPT4 from @OpenAI for our work on @bing. We took this opportunity to document our experience. We're so excited to share our findings. In short: time to face it, the sparks of #AGI have been ignited.
arxiv.org/abs/2303.12712
Yuval Rabani retweeted
And now for an interlude from Transformers taking over the world and the (very unfortunate) Twitter drama:
*The randomized k-server conjecture is false!*
Joint work w. Christian Coester & Yuval Rabani arxiv.org/abs/2211.05753. Picture below is our hard metric space for k-server.
Yuval Rabani retweeted
List of accepted papers to FOCS 2022 now online: focs2022.eecs.berkeley.edu/a…
Yuval Rabani retweeted
AI still has a long way to go .... to me this example is exactly what happened with the whole "sentient discussion": if you prompt with the seed of an answer, the transformer architecture will latch onto this seed. It's really a game of mirrors ...
Yuval Rabani retweeted
So OpenAI got scooped by a Chinese academic lab in a matter of weeks… Claims inference on sub-A100 GPUs possible by specifying smaller batch size, checkpoints only ~10GB model.baai.ac.cn/model-detai…
CogView2: Faster and Better Text-to-Image Generation via Hierarchical Transformers
abs: arxiv.org/abs/2204.14217
CogView2, shows very competitive generation compared to concurrent state-of-the-art DALLE-2, and naturally supports interactive text-guided editing on images.
Yuval Rabani retweeted
Lots of discussion around #LaMDA is missing the point: ofc it's not sentient but the issue is that those systems are so good at mimicking that non-technical people can easily be fooled. As is often the case when topics escape experts, the truth matters less than how it "feels".
Yuval Rabani retweeted
Transformers are changing the world. But how do they learn? And what do they learn?
Our 1st @MSFTResearch ML Foundations team paper proposes a synthetic task, LEGO, to investigate such questions.
Sample insights on Transformers thanks to LEGO below 1/8
arxiv.org/abs/2206.04301
Yuval Rabani retweeted
5000 double pendulums, with a tiny (0.000001 rad) variation on starting angles. Color line connects original neighbours, thickness proportional to how close they remain (firs bifurcation starting at 0:12 provides a good understanding of what's going on).
Intel's 8080 processor, and a bit more history:
Excellent reminder from @adwooldridge that re-running the Seventies won't be all bad: bloomberg.com/opinion/articl…
1
Yuval Rabani retweeted
This thread is mind blowing.
“A still of Kermit The Frog in Blade Runner 2049 (2017)” #dalle
Yuval Rabani retweeted
נמנעתי במתכוון מלקרוא את המאמר של רצף לוי לעומק, די ברור מקריאה שטחית שזה עוד אחד ממליון מאמרי זבל (וכל ביקורת רצינית עלולה להוביל לבריונות כלפי המבקר). בכל זאת, טוב שיש אנשים רציניים שעושים מהמאמר הזה קציצות. @gidmk מראה שהמאמר לא רק חלש טכנית, הוא גם עמוס במניפולוציות.
Replying to @GidMK
17/n This is a classic @callin_bull point - here is the original graph vs my extracted data with one tiny change to the axes (can you spot it?)
Suddenly, it looks like COVID-19 cases are more closely tied to CA calls than vaccines!
I’m happy to have contributed to one of these results. Seb’s take on the broader context is fabulous!
New video! I explain the full story of Set Chasing, how it starts with the Steiner point (most stable selector in convex sets), then goes to Mirror Descent with multiscale entropy, and finally the *evolving entropy* for problems s.a. graph traversal.
youtu.be/FoJeRFOlp48
3
Yuval Rabani retweeted
A massive progress on the KLS conjecture and Bourgain's slicing problem by Klartag and Lehec! 😮 arxiv.org/abs/2203.15551 🎉
Yuval Rabani retweeted
New paper w. Christian Coester & @yrabani, the culmination of my main pandemic project!
We give an optimal* exploration strategy for problems as in the pic below where a robot starts at the source s and tries to get to the target t while room is being revealed ``frame by frame".
Yuval Rabani retweeted
FOCS 2021 begins on Monday, February 7, 2022!
The welcome session is at 9:30am US Eastern Time.
All sessions are virtual.
You may never get another chance to attend FOCS in the wrong year :)
It's not too late to register:
focs2021.cs.colorado.edu/reg…

























