AI is growing incredibly fast! We feel overwhelmed and struggle to keep up with the pace at times. What new tools and features are being released today? Come check out our web https://t.co/rS7g4gtBKc which helps you navigate the state of the art in AI. Follow us @TodaysAIdotai
@CatGodSandHive Thank you for your interest! See our paper starting line 281: we first rank the conditional branches using posterior probabilities P(ยฌ๐ก๐|ยฌassert13). We then flip each ranked branch during code execution to obtain counterfactual values. We add these values to augment the graph
Check out our recent OOPSLA 26 paper: Prosecutor: Bayesian Counterfactual Fault Localization (https://t.co/SDfCyRejOC), Collaborating with University of Southern California. Mukund and I had this idea when we walked back from the ICSE banquet :)
Today, we have one volunteer joining us to build https://t.co/cVTF42lkQ1. We are very excited! Please tell us what features you want to see by click "Request AI tool or service" on our web. Thanks!
We did a fun project of evaluating SOTA video generators on "fantasy" scenarios. The models have challenges on generating implausible actions and video/audio synchronization, but video generation is getting there!
https://t.co/4Q5YvSuJrB
I was adding QR code to an image. I talked to GPT 5 minutes in all types of ways, and the QR code is not scannable. But for Gemini, it worked with one shot, wow
Check out our recent ICLR 26 paper "CodeSense: a Real-World Benchmark and Dataset for Code Semantic Reasoning" https://t.co/bkephe0B2l
Here is the leaderboard: https://t.co/97A0ha0f0s
collaborating with Baishakhi Ray's Group
@baishakhir
@JAldrichPL Lol, you indeed know a lot of details. Upgrade was great, I just went to the market buying some additional memories or a bigger hard drive. GPU fans sometimes can be loud ๐
@JAldrichPL Wow! Always fortunate to have excellent graduate students like this. Congratulations to both of you. I will be looking forward to reading the paper ๐
Today, Dr. Quest from Mayo Clinic gave us a guest lecture on AI in hospitals, ranging from IVF predictions to hospital capacity planning โ two hours and a half talk and discussions, close to 100 slides. We laughed and learned a lot of stuff. Many thanks!
@MingZChen76 Hi Ming, great question!
We compared our soft assertion techniques with traditional fuzzing methods on the 79 programs in GRIST benchmark. We can trigger all the instability using 0.646 sec per program on average, much faster compared to baselines (see Table 6 in our paper).
Here is our FSE 2025 paper on automatically detecting numerical instability in machine earning code. We developed a new approach called "soft assertion" that can automatically learn instability conditions and generate test inputs to trigger instability.
https://t.co/Q7O0X64D3a