Afshin Rahimi retweeted
להתפתחות הזו, אם היא נכונה, יש השלכות חסרות תקדים. למעשה, קיימות רק שתי אפשרויות:
א. המידע נכון. במקרה כזה, החל ממחר מדינת ישראל תצטרך להתחיל לחלק מחדש לאזרחיה מסכות אב"כ וערכות אטרופין, כדי להגן עליהם מפני האיום הכימי שלפתע חזר לחיינו.
ב.המידע אינו נכון, ומדובר בניסיון לנפח מידע מודיעיני כדי לשכנע את הממשל האמריקני לתקוף את איראן, לאחר שהנימוקים האחרים שהוצגו בפניו לא הצליחו להביא אותו להחלטה כזו.
אני מסרב להאמין שממשלת ישראל תהיה מוכנה להפקיר את אזרחיה מול איום כימי ממשי, או לחלופין להטעות במכוון את הממשל האמריקני כדי לגרור אותו למלחמה.
ולכן, אין לי ספק שכבר ממחר תתחיל חלוקה מחודשת של מסכות אב"כ וערכות אטרופין לאזרחי ישראל.
הרי לא ייתכן שמדובר באיום חמור מספיק כדי להצדיק מלחמה, אבל לא חמור מספיק כדי להיערך להגנה על אזרחי ישראל.
Afshin Rahimi retweeted
This is the coolest "application" of LLMs I have seen in a while: instead of using an LLM to solve a problem set, flip the table and ask the student to TEACH the LLM how to solve the problem set. Really lovely.
Thx to Jordan Hoffmann for sharing this :-). youtube.com/watch?v=om7VpIK9…
Afshin Rahimi retweeted
Clearly LLMs must one day run in Space
Step 1 we harden llm.c to pass the NASA code standards and style guides, certifying that the code is super safe, safe enough to run in Space.
en.wikipedia.org/wiki/The_Po… (see the linked PDF)
LLM training/inference in principle should be super safe - it is just one fixed array of floats, and a single, bounded, well-defined loop of dynamics over it. There is no need for memory to grow or shrink in undefined ways, for recursion, or anything like that.
Step 2 we've already sent messages out to Space, for possible consumption by aliens, e.g. see:
Arecibo message, beamed to space:
en.wikipedia.org/wiki/Arecib…
Voyager golden record, attached to probe:
en.wikipedia.org/wiki/Voyage…
The Three Body problem (ok bad example)
But instead of sending any fixed data, we could send the weights of an LLM packaged in the llm.c binary, with instructions for the machine code. The LLM would then "wake up" and interact with the aliens on behalf of the human race. Maybe one day we'll ourselves find LLMs of aliens out there, instead of them directly. Maybe the LLMs will find each other. We'd have to make sure the code is really good, otherwise that would be kind of embarrassing.
:) Step 2 is clearly not a serious proposal it's just fun to think about. Step 1 is a serious proposal as, clearly, LLMs must one day run in Space.
Afshin Rahimi retweeted
🔥Excited to announce our third keynote talk: "Towards Effective NLP Systems: Cultural Norms, Explainability, and Reasoning Enhancement" by Reza Haffari🎇
More details: alta2023.alta.asn.au/program…
Register for #ALTA2023: alta2023.alta.asn.au/registr…
🔥#ALTA2023 Extremely pleased to announce our first two keynote talks:
💫Heng Ji: "SmartBook: An AI Prophetess for Disaster Reporting and Forecasting"
💫Terrence Szymanski: "The Commoditisation of NLP in Industry"
More details: alta2023.alta.asn.au/program…
#NLProc #MelbourneNLP
Afshin Rahimi retweeted
🔥#ALTA2023 Extremely pleased to announce our first two keynote talks:
💫Heng Ji: "SmartBook: An AI Prophetess for Disaster Reporting and Forecasting"
💫Terrence Szymanski: "The Commoditisation of NLP in Industry"
More details: alta2023.alta.asn.au/program…
#NLProc #MelbourneNLP
Afshin Rahimi retweeted
Just published - congratulations team. Automatic Identification of 5C Vaccine Behaviour on Social Media (Sampath Kumar et al., WNUT 2022) @ashrayme @AminathShausan
aclanthology.org/2022.wnut-1…
Deadline extended to 24th August
Working on social media data, medical records, student essays, customer reviews, etc.? #nlproc #socialmedia
Consider submitting a paper (4 or 8 pages, due Aug 19) to the COLING 2022 Workshop on Noisy User-generated Text (W-NUT) -- noisy-text.github.io/2022/
Afshin Rahimi retweeted
aaaand extended a couple of days until Aug 24 ... sorry, a last minute internal decision in light of other relevant #nlproc dates
Afshin Rahimi retweeted
Working on social media data, medical records, student essays, customer reviews, etc.? #nlproc #socialmedia
Consider submitting a paper (4 or 8 pages, due Aug 19) to the COLING 2022 Workshop on Noisy User-generated Text (W-NUT) -- noisy-text.github.io/2022/
Afshin Rahimi retweeted
Replying to @altanlp @barbara_plank
(2) Ben Hutchinson: "Putting NLP Ethics Into Context"
and
(3) Dirk Hovy @dirk_hovy : "More than words – Integrating social factors into language modeling"
Register (free) at:
trybooking.com/events/landin… #NLProc #ALTA2021
Afshin Rahimi retweeted
ALTA 2021 is happening next week! If you are interested in NLP, want to hear some amazing talks and keynote speakers, or want to socialise with NLPers down under, registration is open! #NLProc #ALTA2021 @ashrayme @WLaneguage
We've just released the program schedule for ALTA 2021!
Check it out: alta2021.alta.asn.au/program…
And register to attend (free!) here: alta2021.alta.asn.au/registr…
Afshin Rahimi retweeted
Delighted to announce that the 8th iteration of WNUT will take place at #coling2022 in Gyeongju (dare I say at least partially in person?), co-organised with the inimitable @cocoweixu @alan_ritter and @ashrayme: noisy-text.github.io/
alta2021.alta.asn.au/registr… The 19th Annual Workshop of the Australasian Language Technology Association, registration is free for anyone interested. Keynote speakers: @dirk_hovy @barbara_plank @benhutchinson Hands-on Tutorial: MT&Summarisation. All online #ALTA2021 #NLProc
Class-imbalance and fairness are often studied separately, however, they're intertwined. See our paper and presentation at #EMNLP2021 #nlproc docs.google.com/presentation… arxiv.org/abs/2109.10444 a work with @shivashankar_rs @eltimster @trevorcohn @leafrermann
Submit your NLP/CL/L papers to ALTA2021 (Australasian Language Technology Association), virtual and free for everyone (🙏to sponsors @CSIRO @go1com @DefenceScience @Google) Deadline 1 Oct., alta2021.alta.asn.au/ with fantastic @barbara_plank as one of the keynote speakers.
Afshin Rahimi retweeted
First time ever a language model has been trained directly from raw audio: no text, no ASR. @facebookai's model has been trained on 8000 hours of audiobooks ( just the audio!). Given a short audio prompt, it can continue and generate new speech with its own voice.
Afshin Rahimi retweeted
Good luck to all participating in the 2021 ALTA shared task! Deadline: 22 September 2021. This year's topic: Automatic Grading of Evidence - 10 years later alta.asn.au/events/sharedtas… @altanlp
Afshin Rahimi retweeted
Call for Problems for the 2021 ALTA Shared Task: ALTA is seeking problems to use for the 11th ALTA Shared Task. The closing date for submission of proposals is 4 June.
For questions or to submit a proposal, please send an email to: shared.task at alta.asn.au @diego_ma
Afshin Rahimi retweeted
Website for the WNUT workshop at EMNLP 2021 is now live!
noisy-text.github.io
This is the 7th year we are running the Workshop on User-generated Text. We are glad to see this research community continues to grow. #nlproc
@emnlpmeeting Please RT and submit your work!
Afshin Rahimi retweeted
"Set-to-Sequence Methods in Machine Learning: a Review" - new paper covering this group of methods, w/ Mateusz Jurewicz and a help from @RasmusPagh1 . -> arxiv.org/abs/2103.09656



















