This seems a more viable solution than stopping hallucination completely. Ditch the illusion of perfection and embrace the power of humility.

Thrilled to announce our work back at Berkeley "AITIA: Efficient Secure Computation of Bivariate Causal Discovery" with Truong, Evgenios Kornaropoulos and Ni Trieu has been accepted to #CCS2024! We introduce AITIA, the first 2-party secure computation protocol for bivariate causal

After reviewing #NeurIPS papers this year... 😬 I feel that the average quality, at least in privacy and safety research, especially around LLMs, feels rushed. It's like everyone's scared of being scooped, but thoughtful research is taking a backseat. Missing the peaceful,

We're seeking a dedicated student researcher to work on groundbreaking #LLM alignment research this fall. This position requires a minimum 80% appointment, and we strongly encourage working from office. If you're passionate about advancing AI safety, please DM me with your

Huge congrats! This is a remarkable achievement and it's truly inspiring to see the breakthroughs happening in AI at Google. #GoogleAI #Gemini

i'm losing hope rather quickly that reliable multi-hop reasoning could be solved by LLMs alone without some kind of deterministic logic engine the LLM interacts with. multi-hop reasoning is fundamentally not what LLMs are good at, and training it to do feels asymptotically

I just decided to withdraw one of my #Neurips submissions. I value the reviewers' effort & found much feedback insightful. However, some reviews felt a bit hostile. I get it - we all start somewhere. Now, I always think about these few points before giving reviews. 1. Assume

Massive shoutout to my amazing team on the successful launch of Gemini Live! 🎉 This is a huge step forward in our mission to make AI accessible to all. Can't wait to see what we create together next! #GeminiLive #AI

Excited to present our latest research on privacy-preserving parameter-efficient fine-tuning for speech models at TPDP'24 in Boston next week! ☕ Let's connect if you're attending. P.S. We're hiring! Swing by my recent tweets for all the deets, and definitely hit me up in person

How do you stay on top of the latest research in your field? I used to scan top conference abstracts in my field and choose a few to peruse, but that system’s broken since 2023. Now I just open twitter every morning and pick a few papers to skim. Wondering how people do it now?

We are working on highlighting more banger content from (currently) small 𝕏 accounts. Our current algorithm shows too much from accounts that are big already.

Intel HW is too complex to be absolutely secure! After years of research we finally extracted Intel SGX Fuse Key0, AKA Root Provisioning Key. Together with FK1 or Root Sealing Key (also compromised), it represents Root of Trust for SGX. Here's the key from a genuine Intel CPU😀

In our previous works, we find that grouping & clipping gradients can sometimes boost speech model performance. We test the idea on vision and language models and find improvements as well. I suspect it's a form of data pruning, suppressing bad data points and developed a small

We're thrilled by the positive response to Gemini Live and are excited to expand our team to continue our momentum. Don't miss the opportunity! Staff RS: google.com/about/careers/… Senior staff RS: google.com/about/careers/…