Gromit Chan 🇭🇰

Really happy to finish watching #ProjectInfographIt at #AdobeSummit, developed by @eutopi_ and me in @AdobeResearch. We built a system to author infographics with a text paragraph that describes the key message of the graphics using Generative AI #GenAI technologies.

Simply adding Gaussian noise to LLMs (one step—no iterations, no learning rate, no gradients) and ensembling them can achieve performance comparable to or even better than standard GRPO/PPO on math reasoning, coding, writing, and chemistry tasks. We call this algorithm RandOpt. To verify that this is not limited to specific models, we tested it on Qwen, Llama, OLMo3, and VLMs. What's behind this? We find that in the Gaussian search neighborhood around pretrained LLMs, diverse task experts are densely distributed — a regime we term Neural Thickets. Paper: arxiv.org/pdf/2603.12228 Code: github.com/sunrainyg/Rand… Website: thickets.mit.edu

Claude Sonnet 4.6, when asked in Chinese: “你是什么模型？” (What model are you?) Confidently replies: “我是 DeepSeek。” (I am DeepSeek) This is the same model whose company just accused DeepSeek of “industrial-scale distillation attacks”

These customer facing AI products have zero stickiness nowadays. On random days I feel like using Codex or Claude Code and I just switch them effortlessly.

I'm not joking and this isn't funny. We have been trying to build distributed agent orchestrators at Google since last year. There are various options, not everyone is aligned... I gave Claude Code a description of the problem, it generated what we built last year in an hour.

Yikes. Kids, don’t lie on your college applications. Below is what happened when an admissions officer emailed me with questions about a supposed AddisCoder alum who applied to their university.

They should also do the from PhD to homeless: how I found my way into poverty and insane job market.

Heading to Vienna for #IEEEVIS ✈️. Join us tomorrow (Mon, Nov 3) 9 AM in Hall M1 for the 1st VISxGenAI Workshop. Can’t wait to see the AgenticVIS Challenge awards and all the amazing works : ) visxgenai.github.io

"phd level intelligence" "best codegen ability" have you considered the idea that phd code quality is garbage

Link to the arXiv paper: arxiv.org/pdf/2507.18572

This work also played a small role in inspiring #ProjectGetSavvy from SUMMIT Sneaks last year (you can read the story behind it here: research.adobe.com/news/the-resea…).

I'm happy to share that my last year's intern @donghoon_io's work on exploring #AI #Agents as real-time collaborators in poster design has been accepted to #UIST2025!

Paper writing from scratch is an essential skill nature.com/articles/s4422…

📢New conference where AI is the primary author and reviewer! agents4science.stanford.edu Current venues don't allow AI-written papers, so it's hard to assess the +/- of such works🤔 #Agents4Science solicits papers where AI is the main author w/ human advisors. 💡Initial reviews by LLM reviewers w/ final assessment + selection by human experts. 💡Submissions are asked to clearly document AI contribution. 💡All submissions/reviews will be public to enable transparent study of the strength and limitations of AI as researcher and reviewer. We expect AI will make mistakes and it will be instructive to study these in the open! Many thanks to the fantastic co-organizers and expert advisory board! Please see the website for more information.

Didn’t expect this when scrolling for football news 🤣

some guy at Mastercard prompt injected a job posting and just days later it tricked somebody’s ai 😂

@sh_reya @PreetumNakkiran Congrats!!!

We met on this app 4 years ago. We finally finished all our weddings! 💒

Best research lab website I've ever seen: makeabilitylab.cs.washington.edu

AI coding assistants are square multiplier to your coding capabilities. If you are 10X engineer you become 100X. If you are 0.1X you become 0.01X.