Olivia Wang
@oliviawpy2023
Followers
37
Following
138
Media
12
Statuses
89
CSE PhD student @UCSC. Research interests in Human-Centered #XAI (#hcxai), Neurosymbolic AI & Law (#NeSy). Ex-Software Engineer. Views (and the dog) are my own
Santa Cruz, CA
Joined April 2023
What a week! From Atlanta to Santa Clara, we have presented our work at both the ACS Conference and the BayLearn Symposium! I reunited with my ex-colleague Nina Narodytska, whom I met during my time at VMware, and listened to many, many amazing talks.
1
0
3
That’s why we need trustworthy AI. Not just AI that was designed to achieve SOTA performances on benchmarks, but AI that can be interpreted and understood by human users.
“An AI can pass medical tests yet still reason in unsafe, careless, or biased ways, which makes it untrustworthy for real patients.” at this point. nobody should be surprised.
0
5
17
I cannot believe that it is 2025, and I just came across the awesome work done by @blairstanek and @ben_vandurme on statutory reasoning! I think anyone interested in Neuro-symbolic AI applications in the legal field should read this paper.
0
0
1
I think the meanings of having such gatherings are to find your audience, foster discussions, pass down generational wisdom, and inspire the newer generations of researchers. Thanks for having us ❤️😊 Special thanks to @leilanigilpin and the AIEA lab!
0
0
0
@leilanj I can’t believe I @ the wrong Leilani 😅 and I just discovered it now. Sorry @leilanigilpin
0
0
0
@leilanj Credit: Kahneman, D. (2011). Thinking, fast and slow. macmillan. Kroening, D., & Strichman, O. (2016). Decision procedures, volume 1. Springer. Woof it’s a long thread 😅
1
0
0
I had such a honor mentoring those super talented high school students in the summer, and any universities that they are applying to should consider admitting them! Special thanks to @leilanj and the AIEA lab for their support ❤️❤️
1
0
0
My high school interns, Ryan Bai and Tashvi Bansal are here with me for our presentation. Thanks for making it all the way from California! We are missing Emily Chui, so I am adding a group photo from the summer!
1
0
0
Moreover, by combining the output of each instruction, we could reliably construct a reasoning path for how LLMs reach their final conclusions. We will also be presenting on Thursday at BayLearn as well! Please come say hi if you are coming to BayLearn!
1
0
0
By instructing LLMs to follow the chain-of-decision, we want to test the capability (or limit) of LLM converting fast and intuitive system 1 thinking into deliberate and effortful system 2 thinking, without expensive model training and fine-tuning.
1
0
0
The idea of atomic instructions originate from formal methods and decision procedures, where given a formula, the algorithms should always terminate with a “yes” (satisfiable) or “no” (unsatisfiable) answer. (Kroening & Strichman, 2016)
1
0
0
By curating the list of atomic instructions to follow and verify, our preliminary results show that it can boost the performance of probably the hardest task in logical reasoning- logical fallacy classification.
1
0
0
Chain-of-thought is not explainability, but chain-of-decision is. Today we are presenting our preliminary work at the Twelfth Annual Conference on Advances in Cognitive Systems, “Follow My Lead: Logical Fallacy Classification with Knowledge-Augmented LLMs”.
1
1
1
What does it take to advance AI literacy? Our experts—Virginia Byrne (@VirginiaLByrne) from @MorganStateU, David Broniatowski (@Broniatowski) from @gwuengineering, Hal Daumé III (@haldaume3) from @UofMaryland, and Brandeis Marshall (@csdoctorsister) of @DataedX_—explain how
6
8
8
Is it just me or everyone’s Twitter Avatar just became super blurry?? 😅😅
0
0
0
This is one of the most promising directions to improve RAG systems. It involves combining dynamic retrieval with structured knowledge. It helps to mitigate hallucinations and outdated information, and improves knowledge quality. Pay attention to this one, AI devs!
0
0
1
🚨 BREAKING: 250-year-old Encyclopaedia Britannica is suing Perplexity for copyright infringement, showing what happens when traditional publishers, legal uncertainty, and aggressive AI players clash. Quotes: "Perplexity’s so-called 'answer engine' eliminates users’ clicks on
10
126
402
Hopeless romantics in 2015 majored in English Lit; hopeless romantics in 2025 major in Computer Science 👀👀😬😬
0
0
0
I love this analogy. Being a hopeless romantic, I understand why I chose a PhD instead of my well paid job 🤣
For a PhD, you need to be a romantic at some level. Your papers will get rejected. Your ideas will get scooped. All while you peers flourish. And yes--It will sting. 2023 was one such year for me. Yet I call it my golden year, because that’s when I truly fell in love with my
1
1
3
LLMs just learned how to explain their own thoughts. Not only do they generate answers, they can now describe the internal processes that led to those answers… and get better at it with training. We’re officially entering the era of self-interpretable AI. Models aren’t just
0
0
2