Akshat Gupta
@akshatgupta57
PhD student @UCBerkeley @berkeley_ai | MS @CarnegieMellon | Previously : AI Research @jpmorgan
Our work on knowledge editing got an "Outstanding Paper Award"🏆🏆 at the @RealAAAI KnowFM Workshop!! #AAAI2025 🥳🥳🥳 Congratulations to my amazing co-authors @tom_hartvigsen @_ahmedmalaa @GopalaSpeech
Thrilled to share that our paper on "Norm Growth and Stability Challenges in Sequential Knowledge Editing" has been accepted for an Oral Presentation at the KnowFM workshop @ #AAAI2025 w/ @tom_hartvigsen @_ahmedmalaa @GopalaSpeech More details below (1/n)
Language needs to evolve already and drop the 'er' in 'itinerary.' Just make it 'itinary'—the extra letters are doing nothing for anyone, especially when the shorter alternative is available.
With all the recent discussion around LLM reasoning, I’m awestruck by how insightfully Richard Hamming was asking the same questions about machines and thinking, BUT 30 years ago! youtube.com/watch?v=aq_PLE… youtube.com/watch?v=URp-uq…
Glad you find it useful and thanks for sharing! Paper - arxiv.org/abs/2409.12951
Very well written paper on LayerNorm’s geometry 🧠 Turns out LN explicitly removes a vector’s projection on the uniform direction. But most LLMs naturally learn to be orthogonal to it anyway! 👉 RMSNorm skips this step ✅ Faster ✅ No info loss ✅ Same results paper link below:
I'm excited we've got some some papers accepted to ACL this year on model editing and on scaling laws for quantized LLMs 🎉 Lots of great work from really talented students and collaborators
Wrote a blog about my work on poker + AI (LLMs) with @RichardZ412 and @akshatgupta57. Solvers are not the end of AI integration in poker, and many more positive use case can be explored.
Google’s AI just hit me with a triple negative so hard I forgot my own name - “It’s not uncommon for Mission: Impossible to not have post-credits scenes.”
Very cool paper - "Memorization seems to be happening for prefixes present at the beginning of the context window" arxiv.org/abs/2505.13171
Excited to have two paper accepted at #ACL2025 !🎉🎉 1 Main track and 1 Findings. Papers out on arxiv soon. Big thank you to all my collaborators!
Great to see our work with @DrMayaPetersen, @emrek, @cholmesuk and @mark_vdlaan featured by @MSFTResearch!
Discover how Microsoft Research is leveraging AI to transform research processes, accelerate innovation, and drive growth. Learn more about the strategies and tools empowering researchers here: msft.it/6011SZK8U
I think o3 does better lit review than openai deep research...
Deadline extended! ⏳ The Actionable Interpretability Workshop at #ICML2025 has moved its submission deadline to May 19th. More time to submit your work 🔍🧠✨ Don’t miss out!
#ICLR25 Our work on characterizing alignment between MLP matrices in LLMs and Linear Associative Memories has been accepted for an Oral Presentation at the NFAM workshop. Location : Hall 4 #5 Time : 11 AM (April 27) @GopalaSpeech @berkeley_ai
I’ll be at #ICLR2025 to present our spotlight paper on automatic method to extract LLM characteristics and perform routing! Also I've been recently delving into the RLVR world, so hit me up if you want to chat about model routing or RL for LLM tool-use!
The @OpenAI podcast makes it seem like every iteration of gpt will be 100x its previous iteration. Does it mean we have our first trillion parameter model in #GPT4_5 and #GPT5 will have 10 trillion parameters? Surely that’s got to be it?
Work led by BAIR students @KayloLittlejohn and @CheolJunCho advised by BAIR faculty @GopalaSpeech "...made it possible to synthesize brain signals into speech in close to real-time." dailycal.org/news/campus/re… via @dailycal
1/n) Our latest work is out today in @NatureNeuro! We developed a streaming “brain-to-voice” neuroprosthesis which restores naturalistic, fluent, intelligible speech to a person who has paralysis. nature.com/articles/s4159…