Assistant Professor @Cambridge_Uni. Fellow @Kings_College.
🇺🇸⇔🇬🇧
Joined July 2016
- Tweets540
- Following421
- Followers1.7K
- Likes3.3K
Umang Bhatt retweeted
I’m looking to hire a postdoc researcher. Please see info under News on mlg.eng.cam.ac.uk/adrian
Umang Bhatt retweeted
🫵Are YOU an economist, AI researcher, or otherwise interested in measuring AI’s impact on people and society?
We wrote a paper to make your life easier: we conducted dozens of hours of expert interviews, transcript analysis, and research to identify 9 challenges in AI RCTs/human uplift studies and 9 fixes to make your research more rigorous.
We also mapped each challenge to the type of validity it threatens (construct, internal, external), and its degree of LLM-specificity.
Amidst questions about RSI and the societal implications of frontier AI, we hope this work can support stronger measurement and transparency towards more informed decision-making.
Catch @kevinlwei presenting this work at @AIESConf in Malmö this month. S/o to our coauthors @ShenZhouHong @dbateyko @xave_rg @thecarsonezell @valeriechen_ @umangsbhatt.
Paper: arxiv.org/pdf/2603.11001
PS - before working in AI auditing/evals/governance, I spent five years designing and running causal inference research, including large-scale human-subject RCTs. I’m excited about raising the ceiling at the intersection of these worlds. Check out these comments from a great economist. Hope this work can help others too!
Umang Bhatt retweeted
Today, Carnegie Mellon University and Ken Griffin, founder and CEO of Citadel and founder of Griffin Catalyst, announced a historic $3 billion gift to elevate Carnegie Mellon in Pittsburgh and launch Carnegie Mellon University Miami. #CMUMiami
Learn more: cmu.edu/news/stories/archive…
Run, don’t walk to join Deb @StanfordHAI 🌲
Umang Bhatt retweeted
Yesterday, we opened up @10DowningStreet to the UK’s brilliant AI Olympiad teams from 2025 and 2026.
Exceptional students with such bright futures, making our country proud internationally 🇬🇧
If you are a student, keep an eye on future Olympiads 👉🏽 ukoai.org/
Umang Bhatt retweeted
High quality AI safety communication and transparency is critical. We're hiring for a role on my team, to work closely with @dgrobinson, to supercharge our safety transparency artifacts. This includes system cards, safety research blog posts, and more.
openai.com/careers/safety-tr…
Umang Bhatt retweeted
I am excited to announce that I am joining @perplexity_ai as research lead! We will be doing ambitious paradigm shifting work, advancing the frontiers in the open. If you want to join us in re-imagining continual learning, agent collaboration, and beyond, please reach out!
Umang Bhatt retweeted
AI is making tremendous strides in mathematics. How and where these growing AI capabilities impact mathematical practice, however, is critically a function not just of AI capacities, but whether and where people choose to use them! Not just AI agency, but *human agency,* and how the two interact.
Umang Bhatt retweeted
Games have long served as a testing ground for understanding problem-solving and strategic reasoning in humans and machines, across many fields – from AI, to cognitive science, economics, and more. Yet, in life people are regularly faced not just with repeated instances of the same problem or “system of rule and reward”, but entirely new problems, new systems, new games.
And when encountering new games, people not only need to decide what actions to take but whether a game is even worth playing at all!
How can people think reasonably about new games for the first time, so flexibly and quickly?
We take steps towards this question in our new work, out now in @Nature!
Umang Bhatt retweeted
I've noticed this trend of AI time savings failing to fully manifest and instead being replaced with repetitive work. Technological affordances also require organizational change in order for the savings to benefit workers.
The Bank of Korea just released a report about AI productivity.
Korean workers using generative AI cut task time by 3.8%, about 1.5hours/week on a 40-hour schedule, yet the link between saved time and more work completed was 0.
Only 4.4% of tasks saw time savings above 20%
This creates an AI productivity disconnect, where faster reports can create more reports, faster review can create more review, and saved time can get absorbed by organizational habits instead of becoming higher output.
Umang Bhatt retweeted
As my time in Pittsburgh comes to an end...
I'm excited to share that I will be joining UIUC as an Assistant Professor in fall 2027! I will recruit PhD students in both ECE and CS.
I’m looking for students and postdocs who are excited to continue building towards collaborative AI systems that learn and adapt through interaction.
Umang Bhatt retweeted
🫡 After 7 years, I've just left Google Deepmind to start an AI Safety nonprofit, around Scalable and Human Oversight (i.e. building stronger "judges")! 🇨🇦 And, I'll be at FAccT in Montreal this week! (1/4🧵)
Umang Bhatt retweeted
Clarification (got many msgs about this): I do NOT think that we should ban AI in education, even if these results hold up in larger studies. But we should think very carefully about how we deploy AI and how we align AI with long-term user value (eg learning when not to assist)
🚨📄 New preprint! We find the “boiling the frog” equivalent of AI use. In a series of RCTs, we show that after just 10 min of AI assistance people perform worse and give up more often than those who never used AI.
w Grace Liu @brianchristian Mira Dumbalska and Rachit Dubey 🧵
Umang Bhatt retweeted
🚨New paper 🚨Human uplift studies -- RCTs measuring how AI changes human performance -- inform how we govern, deploy, and understand AI systems and their societal impacts.
My latest piece in Noema on the hidden costs of agentic AI: humans recruited to sense the physical world on an agent's behalf, potentially without their consent. We're all sensors. Link below.
“We like to imagine AI hunting for us, going out into the digital world to retrieve facts, schedule meetings & optimize our lives. But a reversal is underway. We are becoming the gatherers: collecting the offline signals our agents need to continue the hunt.”
—@umangsbhatt
noemamag.com/ai-agents-are-r…
Umang Bhatt retweeted
We introduce epiplexity, a new measure of information that provides a foundation for how to select, generate, or transform data for learning systems. We have been working on this for almost 2 years, and I cannot contain my excitement! 1/7
1/🧵 We are very excited to release our new paper! From Entropy to Epiplexity: Rethinking Information for Computationally Bounded Intelligence
arxiv.org/abs/2601.03220
with amazing team @ShikaiQiu @yidingjiang @Pavel_Izmailov @zicokolter @andrewgwils
Umang Bhatt retweeted
LMArena office is buzzing today. The environment is superlinear. We are constructively interfering, always pushing one another to learn and exceed what we thought were our limits.
The #1 trait that we look for when hiring is "Excellence." Producing work every day that meets the highest technical bar in terms of craftsmanship, performance, and reliability. There are other factors too, but never compromise on excellence. As a consequence, everyone on our team is a deep expert. This is required to build quality evaluations with confidence.
Excellence begets excellence. Nobody wants to work in an environment with a bunch of wet blankets -- it kills the momentum. We're constantly raising the bar, and this is what leads to the feeling of energy. It's rare.
If you're interested in working in this kind of environment, send me a DM. We're always looking for people that can raise the bar one notch higher.
Umang Bhatt retweeted
Hi All! W/ @friederrrr + colleagues, we're studying how people do/don't use AI for formal mathematics. If you're a practicing mathematician, maths student, or otherwise do formal maths, please consider responding to this brief survey for our research! forms.gle/Dv9JbvUvpLeV8ryc7
