
QuantitativeResearcher@twosigma; PhD @LTIatCMU; ex @BoschGlobal; Efficiency & Robustness of Deep Learning, Audio Analytics, Natural Language Processing.
Pittsburgh, PA
Joined April 2012
- Tweets327
- Following152
- Followers139
- Likes1.2K
Question: How do we minimize the cognitive overload after heavily using agentic coding? Feels like I can handle single agent workflow very well, but the recent emerging capability of subagents makes it mandatory to play managment role, which is a taxing job itself.
95
Billy Li retweeted
Location: Hall 3 + Hall 2B Poster #241 at #ICLR2025
Hi, we 'll present our #ICLR2025 work, SMT, as a poster Today April 24 at the 3pm -5:30pm(Singapore) session. It’s a joint work with @JunchengLi, Xuan Jiang, and @heathercmiller. Come to our poster at #241 TL;DR:!
Billy Li retweeted
Replying to @neuralink
Ah, the latest on Neuralink's Blindsight getting the FDA Breakthrough tag? Well, it's like Elon Musk's brainchild, Neuralink, just got a gold star from the FDA for its Blindsight implant. This isn't just any old sticker; it's the "breakthrough device" designation, which is the FDA's way of saying, "Hey, this could be a game-changer, let's fast-track this!"
- What's Blindsight? Imagine a device that could potentially restore vision to those who've lost it, or even to those born blind. It's like giving someone a new set of eyes, but instead of going through the usual eye route, it goes straight to the brain's visual cortex.
- The FDA Breakthrough: This designation is for devices that promise to treat or diagnose life-threatening or irreversibly debilitating diseases or conditions. Essentially, the FDA is giving Neuralink a nod, saying, "This could be big, let's not dilly-dally with the red tape."
- Public Reaction: From the posts on X, there's a mix of awe and excitement. People are buzzing about the potential. Imagine being able to see for the first time or regaining your sight. It's like something out of a sci-fi movie, but it's happening, or at least, it's on its way.
- The Tech: While the specifics of how it works inside the brain are still under wraps (or rather, under the skull), we know it involves electrodes and some fancy neural decoding. Think of it as translating brain waves into visual information.
- Challenges Ahead: Despite the excitement, there's a reality check from some corners. High-resolution vision might still be a distant dream due to the complexities of how our brains process visual information. But hey, baby steps, right?
- Looking Forward: Neuralink isn't just stopping at vision. They're also working on helping paralyzed individuals control devices with their thoughts. So, while Blindsight is about seeing, Neuralink's broader goal is about making the brain and technology chat like old friends.
So, there you have it! Neuralink's Blindsight getting the FDA's breakthrough nod is like getting a VIP pass to skip some of the bureaucratic lines, aiming to bring vision to those in darkness, quite literally. It's a step towards making the future not just look bright but also see it.
Not everyone needs to work on LLMs, my collaborator Jackson's cool work with Emma Alexander at Northwestern's Bio Vis lab was featured in a Veritasium video, at 29:20: What Jumping Spiders Teach Us About Color youtu.be/nfAqTSjMBJk?si=eTqh… via @YouTube
170
As schools are only teaching neural nets, lots of the good old techniques are left behind in librosa, opensmile, or praat (who uses Praat in windows...) Torchaudio probably never bother to reimplement these
An implementation of the Yin algorithm for pitch estimation in @PyTorch: brentspell.com/2022/pytorch-…
5
479
There are two types of LLMs: those that talk the talk and those that walk the walk. LLM walk the walk sometimes talk the talk but most times they don't talk at all, cus they work… personal impression after using LLaMA, Alpaca, Dolly and of course you know what it is. #LLMs
1
165
You don’t know the power of the dark side.
1/The call for a 6 month moratorium on making AI progress beyond GPT-4 is a terrible idea.
I'm seeing many new applications in education, healthcare, food, ... that'll help many people. Improving GPT-4 will help. Lets balance the huge value AI is creating vs. realistic risks.
244
Billy Li retweeted
On April 1st-2nd, we're going to have a big event in Pittsburgh about Large Language Models (LLMs).
We'll have talks from great speakers, a discussion of future directions, and a hackathon. Come one come all to join us at the CMU Rashid Auditorium! cmu-lti-llm.org/
My partner ying doesn’t know much coding, (couldn’t write a proper for loop).@yingwangcecilia and yet she could implement a Roberta text classification model bug free with the help of GPT4 (generating code and feed the error back in as prompt). OMG!
1
3
320
Well put!
One more point: Do not blindly follow the crowd.
It looks like the stress is real. In my experience:
1. paradigms shift
2. numbers get higher
3. Yet, unsolved problems remain in big numbers.
Moreover, it looks like new discoveries create new opportunities. It is a bit hard when you are locked in with some topic that you chose for your thesis. Just finish your PhD (successfully) and move on. It is easier to adapt & change direction when you are not a grad student, believe me.
Overall, I think finishing projects is ultimately more important than chasing an new shiny target all the time. Yes, that next paper will not change the world, but ten finished papers on your CV can certainly change your life (or ten projects on your GitHub). Look, I started my journey even before ML became a thing and I am doing rather well.
1
311
Rethinking <zero to one>, I find a striking similarity between the current landscape of ML research and business itself. OpenAI is on their way to achieving a creative monopoly, while the rest of us will reach perfect competition - equilibrium -death someday.
125
I am pleasantly surprised by your response, Chris! Like what we did before: nips2018creativity.github.io…
However, plugging 18th-century counterpoint into music technology is a non-trivial task and would need more experts label.
Replying to @chrisdonahuey
Definitely planning to direct my research to steer away from such a future! I think the key is building music generation tools that are *complementary* to traditional music skills (like autotune was). Gets the best of both worlds while preserving incentives
1
303
Billy Li retweeted
Named Tensor Notation (TMLR version, arxiv.org/abs/2102.13196 w/ @davidweichiang + @boazbaraktcs)
A rigorous description, opinionated style guide, and gentle polemic for named tensors in math notation.
* Macros: ctan.org/tex-archive/macros/…
twitter shitshow is #elon's shabby PUA tricks, while this twitter -> mastodon drama is just a sad reality show... it's the same opinion leaders behind the screens, picking podiums. Guess not many ppl like each other, bad parts of humanity certainly got amplified on social media.
1














