DeepMind’s Demis Hassabis is #AI’s grandmaster https://
vox.com/future-perfect
/23375219/future-perfect-50-demis-hassabis-deepmind
…
RESEARCH
-
Demis Hassabis Named AI’s Grandmaster by Vox
By
–
-
Meta AI Columbia U Improve Dialogue Models with Response Refinement
By
–
Meta AI & Columbia U ‘Squeeze the Juice’ to Turn Bad Responses into Good Labels and Boost Dialogue Model Performance https://
syncedreview.com/2022/11/01/met
a-ai-columbia-u-squeeze-the-juice-to-turn-bad-responses-into-good-labels-and-boost-dialogue-model-performance/
… -
Groq et Argonne collaborent pour optimiser les réacteurs de fusion
By
–
We're proud to be working with our customer, @argonne
, and delivering low latency and deterministic performance–both mission-critical to #fusionenergy reactor operations and stability. -
Fei-Fei Li discusses AI with human perspective at MIT
By
–
I still love this building after all these years @MIT_CSAIL Excited to be back; and giving a talk tomorrow https://
calendar.mit.edu/event/dertouzo
s_distinguished_lecture_fei-fei_li_what_we_see_and_what_we_value_ai_with_a_human_perspective#.Y2GPDqQpAWM
… -
Adversarial Robustness Learning Theory Paper Review Process Critique
By
–
"Adversarially Robust Learning with Tolerance" by @ashtiani_hassan
, @OneBigOh
, Urner. 2 +ve reviews, 1 -ve, saying "I implemented the method and it didn't work." But this is a theory paper… Another -ve review 4 days pre-decision, no chance to respond -
NeurIPS 2022 Rejected Papers Now Public for Review
By
–
#NeurIPS2022 papers and reviews are now public, including "Rejected Papers that Opted In for Public Release." Please share your favourite rejected papers (by you or others) in this thread! Show us tough but quality reviews, how reviewing is broken, etc. https://
openreview.net/group?id=NeurI
PS.cc/2022/Conference#rejected-papers-opted-in-public
… -
Google Scholar pricing and Semantic Scholar as billionaire-funded alternative
By
–
It would actually be more comforting if Google charged something reasonable to institutions for Scholar. Interestingly, competitor @SemanticScholar appears to be the result of a billionaire's endowment (the Allen Institute for AI), so maybe it represents a long-term alternative.
-
Passive Optical Components Reduce Client Device Energy Consumption
By
–
An interesting way to reduce energy use on client devices by doing more work via passive optical components. // @MIT_CSAIL // #AI #machinelearning #deeplearning #photonics
-
6.4M AI-Generated Images Map Reveals Human Interests
By
–
Worth exploring: a map of 6.4M million images generated by AI, grouped by what the AI considers to similar images.
— Ethan Mollick (@emollick) 1 novembre 2022
Why you might want to explore it: it shows you what kinds of images humans are interested in; it teaches new prompts; & has surprising art.https://t.co/t1oUai8HJNWorth exploring: a map of 6.4M million images generated by AI, grouped by what the AI considers to similar images. Why you might want to explore it: it shows you what kinds of images humans are interested in; it teaches new prompts; & has surprising art.
-
Standardized Testing Fails to Differentiate Among Thousands of Students
By
–
There is a failure of standardized testing as it cannot distinguish between 10,000+ students. The SAT and GRE are too easy, especially the math sections.