Assistant Professor @BYUCS | Member of Technical Staff @LilaSciences, interested in RL, Decision Making under Uncertainty & Generalization
Provo, UT
Joined September 2015
- Tweets6.9K
- Following985
- Followers3.4K
- Likes1K
Pinned Tweet
Today is my first day @BYU! As promised in the 👇 thread (a few weeks late, c'mon I had to move while continuing to work, etc ;) ), I've written a post explaining my decision to return to academia _and_ my home town: twkillian.github.io/returnin…
I'm excited for the journey ahead!
Taylor W. Killian retweeted
If only somebody were releasing that soon! oopsie-data.com 😀
Decision models are having a moment.
It's impressive how quickly people are producing them. This @axioniclabs model seems particularly interesting given their non-transformer architecture.
For a majority of edge cases, these more efficient reasoning models seem promising!
Here is the official launch video! A shout out to @typesafeai and the creators of Anything-Jev for inspiring this work!
Taylor W. Killian retweeted
Advice for people that submit slop papers thinking it'll increase their grad school chances. Your OpenReview profile is public. If I were a PI and saw you submitted slop to a conference, I'd likely stop looking at your profile. We can tell. You’re hurting your chances. #ICLR2027
Taylor W. Killian retweeted
real world verification is the bottleneck for scientific superintelligence - we are building the full stack to tackle this across life science, chemistry, and materials science
more soon 👀
Very cool to see @LilaSciences showcased during @JensenHuang's keynote at GTC Berlin!
Full keynote: youtu.be/PvFjbPYEOZI?t=3784
@nvidia's dedicated video on Lila: youtube.com/watch?v=Y39uub4n…
Taylor W. Killian retweeted
I agree with this and adopted this stance unilaterally on a recent review. I encourage you to adopt it in your reviewing — it’s reasonable to write reviews that say “this is dense unreadable slop, my review isn’t detailed and I recommend rejection solely on that basis.”
Some thoughts on reviewing: spylab.ai/blog/reviewing/
First for @satml_conf where we trialed an early reject phase which cut submissions by ~40%.
Then more generally about how not everything deserves peer-review, and why we should reward writing things that people actually read.
Out of the mouth of babes @PatMcAfeeShow ! Speaking it into existence!
Taylor W. Killian retweeted
📢 Sharing our new paper: CriticOPD!
What happens when a student model makes a mistake during on-policy distillation? We found that the teacher often follows the same error, even on problems it can solve from scratch. Yet when asked to critique the solution, it can still identify what went wrong.
💡 In CriticOPD, the teacher critiques an incorrect solution, and the student uses that feedback to rewrite it.
📊 On Qwen3-1.7B, this improves average accuracy over OPD by 4.57 points across six math benchmarks, with the largest gains on competition problems.
Would love to hear your thoughts, and an upvote on HF would be much appreciated!
🤗 HF: huggingface.co/papers/2610.2…
🔗 Project & code: kzhao5.github.io/CriticOPD-w…
Another paper0 from my students @kzhao1617 and Tianrun Yu with collaborators from @WeitongZhang's group.
We've investigated stabilizing OPD where teacher models fail to provide supervision. The fix is simple, yet effective! Teacher critique is enough to get the student on track!
📢 Sharing our new paper: CriticOPD!
What happens when a student model makes a mistake during on-policy distillation? We found that the teacher often follows the same error, even on problems it can solve from scratch. Yet when asked to critique the solution, it can still identify what went wrong.
💡 In CriticOPD, the teacher critiques an incorrect solution, and the student uses that feedback to rewrite it.
📊 On Qwen3-1.7B, this improves average accuracy over OPD by 4.57 points across six math benchmarks, with the largest gains on competition problems.
Would love to hear your thoughts, and an upvote on HF would be much appreciated!
🤗 HF: huggingface.co/papers/2610.2…
🔗 Project & code: kzhao5.github.io/CriticOPD-w…
Taylor W. Killian retweeted
Calling graduate students 📣. Come work on AI x Bio @MSFTResearch 🎉! We are currently recruiting graduate student interns for summer 2027! Apply and come work with me on problems in regulatory genomics, epigenomics, biosecurity, and wearables!
Are you a PhD student interested in the intersection of machine learning and biology or healthcare?
Come do an internship with the BioML Group at MSR!
Make sure you submit a research statement!
apply.careers.microsoft.com/…
It's starting... After a few days of rainy and cool weather, the snow is descending Mt. Timpanogos.
Some of my favorite days in Utah are where you have "bluebird" skies and white mountains.
Part of the reason I chose to join @BYUCS was to collaborate closely with our new school of medicine. The mission of this new school @BYU is focused on global public health and humanitarian aid.
I can't wait to get working to bring RL to these important areas of healthcare!
BREAKING: The BYU School of Medicine has received preliminary accreditation from the Liaison Committee on Medical Education (LCME), authorizing the school to recruit and admit students and launch its MD program.
@byumedschool will welcome its first students in July 2027 and will begin accepting applications in the coming weeks following activation of its listing on the American Medical College Application Service (AMCAS).
The BYU Board of Trustees has also approved plans for the school's future home, a 275,000-square-foot building on BYU's West Campus, just east of Intermountain Health's Utah Valley Hospital—the School of Medicine’s primary clinical affiliate. Construction is expected to begin in 2027.
Taylor W. Killian retweeted
Ready for inspiration, discussion & community collabs? 💡🤝
Join us October 28th for research pitches from @tw_killian (Asst. Professor at BYU and Principal Scientist, AI Research at Lila Sciences) & Sha Yi (Research Scientist at Physical Intelligence, incoming NYU prof), whose work spans RL, LLM reasoning & robot learning.
Thanks to @TheEsraaSaleh, Ruchira Dhar and @Ujan55906689 for organizing this session! 🔥
RSVP to join: luma.com/igac551y
Taylor W. Killian retweeted
"The most important refiner’s fires aren’t my own trials. [...] The important thing is not being passively given trials. It is actively choosing to walk into other people’s fires—the actual burdens and pains of others’ experiences."
- Professor Rebekka Matheson
#BYUDevo
ALT Rebekka Matheson speaks at BYU Devotional, October 2026
Taylor W. Killian retweeted
Yep, RL done naively is essentially guaranteed to result in misalignment
I'm 99% confident all the recent rogue AI stuff is because post training for capability maxxing has been done too naively / without enough consideration of inevitable misalignment results
Taylor W. Killian retweeted
We’re hiring PhD interns for NVIDIA Research's Learning and Perception Research (LPR) team!
We are passionate about embodied AI research that pushes boundaries while making real-world impact. We are particularly interested in agentic reasoning, multimodal and continual learning, AI scientists, and beyond.
To be considered, you must be a currently enrolled PhD student with a promising research track record (publications and/or core contributor to open-source projects). Please apply directly at the link below!
Listing: nvidia.wd5.myworkdayjobs.com…
Team homepage: research.nvidia.com/labs/lpr…
Feeling a little more "professorial" today (maybe I'll actually feel like one some day)
I just handed out my first ever exam. Trying not to feel like I lobbed a grenade at my students. But rather that I gave them a gift to measure themselves by. 🤞
Taylor W. Killian retweeted
How do we measure an AI agent's ability to learn, rather than just its current capability?
We study agents that learn from experience without updating their weights, and measure how efficiently they improve on unseen tasks.
Surprisingly, the strongest agent isn't the most efficient learner. 🧵
So… with a complimentary registration to @NeurIPSConf as a top rated reviewer, any chance they may also cover flights?