BREAKING: @SouthwestAir Southwest Airlines Flight 2094 from Nashville to Fort Lauderdale, Florida was forced to divert to Atlanta late last night after a an Arabic looking Muslim passenger onboard the plane threatened to blow the plane up with a bomb! You can see the SWAT team apprehend the Muslim passenger while terrified passengers were forced to put their hands up. Zero media coverage!!!! Share this everywhere!! DEPORT ALL MUSLIMS FROM AMERICA!! If you were on this flight, please DM me. @DHSgov @FBIDirectorKash @RealTomHoman @POTUS

Check out my latest paper review: SFT Memorizes, RL Generalizes linkedin.com/pulse/paper-rev… with @MikeE_3_14

Programming languages like Python (and even C) were made for humans. When LLMs write the code, why stick with human optimized syntax? Go to the source: Assembly might be the real efficiency go to weapon of choice. wired.com/story/programming-…

When reading AI reasoning text (aka CoT), we form a narrative about the underlying computation process, which we take as a transparent explanation of model behavior. But what if our narrative is wrong? We measure that and find it usually is. Now on arXiv: arxiv.org/abs/2508.16599

Producing reasoning texts boosts the capabilities of AI models, but do we humans correctly understand these texts? Our latest research suggests that we do not. This highlights a new angle on the "Are they transparent?" debate: they might be, but we misinterpret them. 🧵

Nice - my AI startup school talk is now up! Chapters: 0:00 Imo fair to say that software is changing quite fundamentally again. LLMs are a new kind of computer, and you program them *in English*. Hence I think they are well deserving of a major version upgrade in terms of software. 6:06 LLMs have properties of utilities, of fabs, and of operating systems => New LLM OS, fabbed by labs, and distributed like utilities (for now). Many historical analogies apply - imo we are computing circa ~1960s. 14:39 LLM psychology: LLMs = "people spirits", stochastic simulations of people, where the simulator is an autoregressive Transformer. Since they are trained on human data, they have a kind of emergent psychology, and are simultaneously superhuman in some ways, but also fallible in many others. Given this, how do we productively work with them hand in hand? Switching gears to opportunities... 18:16 LLMs are "people spirits" => can build partially autonomous products. 29:05 LLMs are programmed in English => make software highly accessible! (yes, vibe coding) 33:36 LLMs are new primary consumer/manipulator of digital information (adding to GUIs/humans and APIs/programs) => Build for agents! Thank you again for the invite @ycombinator and congrats again on an awesome events! I'll post some links/references in the reply.

Interesting read: Potential future research directions in AI for software engineering openreview.net/pdf?id=RuLsq4… @koushik77

Wanna get a birds-eye view of the topics and trends in #EMNLP2024? check out the link in @UrikaUri 's thread, presenting navigable topics and subtopics, created by his knowledge-navigator literature sense-making system (which will also be presented in EMNLP)

Asaf @asafam is presenting his TACL work on non-programmers text-to-code benchmark , a project lead by @rtsarfaty

Curious about LLMs generating code and #text2code models? Join me today at Poster Session 2 to discover NL Programming—a new task for coding in plain, simple language! ✨📸 #ACL2024NLP #ACL2024

AI Johnny Cash sings "Barbie Girl"

I'd heard that GPT-4's image analysis feature wasn't available to the public because it could be used to break Captcha. Turns out it's true: The new Bing can break captcha, despite saying it won't:

We present MusicGen: A simple and controllable music generation model. MusicGen can be prompted by both text and melody. We release code (MIT) and models (CC-BY NC) for open research, reproducibility, and for the music community: github.com/facebookresearch/…

Text to image prompt ambiguity at its best 😆

שמח שלא הלכתי לכנס עם סאם אלטמן, לא שקיבלתי הזמנה ממישהו 😆 אבל גם לא ראיתי את ההקלטה. בכל מקרה, במקום להתעמק, ללמוד ולהבין פתחו את זה לציבור, לאנשים לא ממש רלוונטיים שהפכו את זה לארוע דיסטופי מעין סוג של שבתרבות פופולארי כזה. כמה שטויות. F

איזה בזבוז זה לקחת אנשים שבוודאות יודעים דברים, ובמקום להתמקד בזה, לתת להם ללהג על מה הם חושבים/רוצים יהיה באיזו פנטזיה עתידית ילדותית ועל הדרך שבה בינה מלאכותית מושלמת תוכל לעזור לנו לפתור את משבר האקלים.

Morphological inflection is a fundamental nlp task, where we get a lemma and a set of features, and predict the respective inflected word-form. With >100 languages with labelled data in @unimorph_ , and with avg accuracy>90 over all languages, it is considered **solved** >>