Obed Junias (@obed_junias) header
Obed Junias (@obed_junias) profile picture

Seeking PhD Positions in AI/NLP for F'27 | Research @cuboulder | Reasoning and Understanding | HCI/HAI | AI Safety

Boulder, CO
Joined October 2013
Photo by Obed Junias (@obed_junias)Photo by Obed Junias (@obed_junias)Photo by Obed Junias (@obed_junias)
Obed Junias retweeted
The survey is also part of an upcoming handbook on interpretability for AI safety. We've developed an interactive companion with an explorer of 51 methods, 3 demos, and 19 hands-on notebooks. More to follow on this! The material has already informed teaching at @EdinburghNLP and will form the basis of my Interpretability for AI Safety course at @aims_oxford in Hilary term 2027 and with more to follow! If you're teaching interpretability and would like lecture slides, notebooks, or an early copy of the handbook, please reach out! 9/ Paper: alphaxiv.org/abs/2611.08438 arxiv.org/abs/2611.08438 Website: tsglab.github.io/mechinterp-…
1
25
2
119
15,448
Obed Junias retweeted
After supervising 20+ papers, I have highly opinionated views on writing great ML papers. When I entered the field I found this all frustratingly opaque So I wrote a guide on turning research into high-quality papers with scientific integrity! Hopefully still useful for NeurIPS
25
292
16
2,700
348,037
I feel so conflicted, on one hand I really love AI in almost all of its applications; on the other hand, I really miss the pre AI world and all of the high focused learning I had to do — read through papers, try and recreate things manually, go through details of everything…
7
2
99
3,140
Bury me with my unread PDFs.
1
3
21
1,055
Come talk to @obed_junias about our work on logical composition in common sense reasoning at the poster session on the 4th floor right now! #ACL2026 #ACL2026NLP
1
8
511
I'll be presenting our #ACL2026 paper, LOGICAL-COMMONSENSEQA, on Monday, July 6th at 9:00 AM in San Diego. I'm currently exploring PhD opportunities and would be glad to connect with faculty, students, and researchers at ACL.
1
2
372
In this work, we study how well LLMs reason over combinations of commonsense statements, not just single facts, but statements combined using AND, OR, and NEITHER/NOR.
1
137
Our results show that negation-based reasoning remains especially challenging for current models, pointing to a meaningful compositional gap. Come say hi if you'll be there! #ACL2026 #NLProc
133
Obed Junias retweeted
Boulder NLP has 19 papers accepted to #ACL2026 across the main conference, findings and workshops! 🚀 We'll be presenting work on morphology, machine translation, narratives, low-resource languages, CSS, HCI, LLMs, fairness/bias, and more. See you in San Diego! 🌊 ☀️ 🌮
3
1
12
1,537
Obed Junias retweeted
When Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT. Anthropic estimated that this would affect approximately 0.03% of traffic.
95
152
270
1,430
1,344,015
Another reason why U.S. PhD programs, and North American PhD programs in general, are often considered superior in quality to many PhD programs in other parts of the world, despite their longer duration, is that you cannot avoid developing a thorough understanding of the fundamentals. You must pass comprehensive qualifying exams, whether written or oral, and earn course credits to demonstrate that you are grounded in the basics. There are no exemptions.
There was a lady who started her PhD a year ahead of me. Eventually, she went for her qualifying exam. After her 30 mins presentation, she was questioned by the committee for another 1 hour and 30 minutes. One of the questions she was asked was: Where do transcription and translation take place in the cell? She missed the question completely. I think the committee was surprised because that was considered very basic science. At the end of the exam, she was asked to assess her own performance, which is a common practice. She rated herself very highly, and I think that became another concern for the committee because her performance had been considered poor. They probably expected a more honest self-evaluation. Sadly, she could not continue with the program after that. That was the end of her PhD journey there. But beyond her experience, I learned something important very quickly: never neglect the basics of science. It is easy to become so focused on complex ideas, advanced techniques, and “big” concepts that you overlook the simple foundations that hold everything together. After hearing that story, I intentionally went back to revisit my fundamentals. I downloaded materials, watched YouTube videos, and even refreshed my knowledge of statistics. Mind you, I had earned a distinction in Biostatistics…😁 Yet, I still realized there was more depth and clarity I needed. And interestingly, statistics was part of the questions I was eventually asked during my own exam. I think many people in science and academia become overly fascinated with complexity while underestimating the importance of truly understanding the basics. But the truth is, good science should be simple enough to explain clearly and understand deeply. We must pay more attention to the fundamentals because they are what truly make science solid.
38
118
25
1,566
510,597
Obed Junias retweeted
Existential risk from AGI is one of the most important issues right now. If you want to understand more about the topic, and technical work that helps, the BlueDot course is a great way to learn! I did a version of it several years ago and it was helpful. Highly recommended
Want to fast-track your technical AI safety career? Apply to our course by 26th April! Our alumni work at Anthropic, Google DeepMind, UK AISI, Apollo, Redwood, MATS and more. We have given top talent introductions to top orgs over $500k in grants to build portfolios and build impactful careers. The course is facilitated by experts to help you navigate the safety landscape and transition into impactful safety work and it is widely recognised by major AI safety orgs and fellowships. You can either do the intensive version over 6 days or part-time over 6 weeks to fit around your job! 🔗bluedot.org/courses/technica…
8
10
189
27,836
Obed Junias retweeted
Don't idolize anyone. We're all just flawed humans.
1,145
3,955
323
23,328
1,417,570
REMINDER!! No job should come at the cost of your sleep or health. You’re one of many to your employer, don’t overwork or stress yourself unnecessarily. After all, it's about you, family, and friends. – at Boise, ID
10
25
390
24,054
Obed Junias retweeted
I'm very familiar with both these places. Trying every day to maximize time spent in the second one. If you're in the first today, hang in there ❤
409
469
68
7,604
470,554
Obed Junias retweeted
POV: 4 stages of friend groups in college
91
325
75
4,431
215,601