Shreshth Malik

Reminder - PhD applications for OATML are now open The first funding deadline is December 2 - candidates interested in developing Bayesian deep learning methodology, applications of ML, AI security, and understanding ML methodology are encouraged to apply More info: oatml.cs.ox.ac.uk/apply.html

I'm at @ICCVConference and will be giving a keynote talk tomorrow at the @SaFeMMAI workshop at 10:55 (room 308 B) Drop by to say hi! I'm also looking for PhD students and postdocs to join @OATML_Oxford - happy to chat during the conference. Email: yarin@cs.ox.ac.uk

⚠️Beware: Your AI assistant could be hijacked just by encountering a malicious image online! Our latest research exposes critical security risks in AI assistants. An attacker can hijack them by simply posting an image on social media and waiting for it to be captured. [1/6] 🧵

After thousands of microscope hours and several years of painstaking optimisation, @glynnca93 and I are very excited to share our latest manuscript. We outline our approach to investigate the molecular organisation within hippocampus tissue using cryo-ET.

Is your LLM hallucinating? 👻 Our @Nature paper shows how to detect when an LLM is making things up. A 'confabulating' LLM answers with inconsistent meanings when re-asked the same question. We use this to estimate uncertainty and detect confabulations. Learn more 🧵👇 1/

We're looking for 2 postdocs to join @OATML_Oxford, working on Foundational AI Safety. You’ll lead and contribute to projects aimed at developing principled and practical safe AI methods which would be used in real systems eg in the medical domain. Closing date 12th June Links>>

Out in Science today: In our paper, we describe extreme AI risks and concrete actions to manage them, including tech R&D and governance. “For AI to be a boon, we must reorient; pushing AI capabilities alone is not enough.”

We’re organising a virtual open day! 📅 4-5:30 PM GMT on 22 Nov 2023 Anyone interested in joining @OATML_Oxford 🥣🌾for a PhD is welcome! Come to meet OATMLers, get a feel for our group, and ask any questions you may have! ✏️ Sign Up: forms.gle/MSMwp3wvUR9Cnc…

Excited to share our new work! 📚arxiv.org/abs/2310.16410 Most interpretability research focuses on understanding which human concepts AI systems encode. We ask a different question: Do AI systems encode superhuman concepts, and can we use them to further human knowledge?🧵(1/12)

FDL-X was a resounding success with compelling results for each team and an integrated ML pipeline that we can't wait to show off 🚀 Tune in to FDL-X live on 9/20 for a live broadcast of this year's results docs.google.com/forms/d/e/1FAI…

Happy to share that I'll be helping the UK taskforce as director of research (together with @DavidSKrueger). We're heavily recruiting - if you have technical expertise and want to work on Frontier Models (LLMs, generative AI), please read here twitter.com/soundboy/statu…

Fully Funded 4-year Doctoral Studentship Joint with Five AI and the EPSRC CDT in Autonomous Intelligent Machines & Systems (AIMS) - Starting October 2023. Further details can be found here: aims.robots.ox.ac.uk/study/industry…

at #icml today? come chat with us about a new method to sample batches for active learning! SPGIM: spigmworkshop.github.io Poster Session: 12-1pm and 4-5pm 📄 arxiv.org/abs/2306.15058 with @ShreshthMalik Salem Lahlou @JainMoksh Kolya Malkin @TristanDeleu Yoshua Bengio @yaringal

Excited to be working on "things moving in space" with @FDL_AI @NASA this summer #explainitlikeim5 #FDLX #helio

A colleague called my attention to Tom Everitt's "Towards Causal Foundations of Safe AGI" alignmentforum.org/s/pcdHisDEGLbx… alignmentforum.org/posts/Qi77Tu3e…. A good read!

To be presented at the Structured Probabilistic Inference & Generative Modeling Workshop @ ICML, Hawaii 2023! Work alongside Salem Lahlou, @anndvision @JainMoksh, Kolya Malkin, @TristanDeleu, Yoshua Bengio and @yaringal - a great collab between @OATML_Oxford and @Mila_Quebec

Tired of greedy approximations for constructing batches in active learning? Introducing BatchGFN - an amortized sampler that chooses data points to label from the *true* (arbitrary) batch objective cheaply at inference time ✅ arxiv.org/pdf/2306.15058…

Today, we’re starting a blog post sequence explaining how causality underpins many critical aspects of safe AGI, including agency, incentives, misspecification, generalisation, fairness, and corrigibility. We explain past work and open questions. tinyurl.com/caus-safe-agi

Tricky alien worlds easier to find when humans and machines team up trib.al/pfixDcV

@letsraave @She__Fi @jessicapointing

It’s @She__Fi season, so nominate your favorite women in web3 coming to Devcon Bogota to get a rAAVE ticket! As simple as that 🎟 ✨ 👻 🫶🏼 ♥️