Full Width [alt+shift+f] Shortcuts [alt+shift+k]
Sign Up [alt+shift+s] Log In [alt+shift+l]
6
Lately, I've been experimenting with interfaces for large language models (LLMs) in my free time. The fruit of this labor is something I'm calling "curie," an exploratory and sense-making tool designed to navigate complex topics. 0:00 0:34 1× the limitations of
a year ago

More from Apoorva Srinivasan

diffusion models for protein generation

Introduction Proteins are nature's versatile nanomachines— they have evolved to perform virtually every important task in living systems. While nature has produced an incredible range of protein functions, these represent only a tiny fraction of what's possible in the protein universe. Evolution has only explored

a month ago 38 votes
a review on protein language models

Protein “language” is a lot like human language. Given the similarities, researchers have been building and training language models on protein sequence data, replicating the success seen in other domains, with profound implications. In this post, I will explore how transformer models have been applied to protein data

8 months ago 8 votes
what's happened since the human genome project

When the human genome project was deemed “complete” in 2003, it was met with incredible fanfare. The entire project leading up to that moment had all the drama to keep its audience enthralled. Fierce rivalry between a public and private institution, multiple countries involved, 3.5 billion dollars

over a year ago 5 votes
niceR code with functional programming

At the end of this blog post, you will be able to: Describe functional programming concepts Write functional programming code using purrr package in R If you are anything like me, you probably focused primarily on learning statistics, machine learning and programming on a smaller scale early on in your

over a year ago 7 votes

More in science

An update, + a paper as a fun distraction

My post last week clearly stimulated some discussion.  I know people don't come here for political news, but as a professional scientist it's hard to ignore the chaotic present situation, so here are some things to read, before I talk about a fun paper: Science reports on what is happening with NSF.  The short version: As of Friday afternoon, panels are delayed and funds (salary) are still not accessible for NSF postdoctoral fellows.  Here is NPR's take. As of Friday afternoon, there is a new court order that specifically names the agency heads (including the NSF director), saying to disburse already approved funds according to statute.   Looks like on this and a variety of other issues, we will see whether court orders actually compel actions anymore. Now to distract ourselves with dreams of the future, this paper was published in Nature Photonics, measuring radiation pressure exerted by a laser on a 50 nm thick silicon nitride membrane.  The motivation is a grand one:  using laser-powered light sails to propel interstellar probes up to a decent fraction (say 10% or more) of the velocity of light.  It's easy to sketch out the basic idea on a napkin, and it has been considered seriously for decades (see this 1984 paper).  Imagine a reflective sail say 10 m\(^{2}\) and 100 nm thick.  When photons at normal incidence bounce from a reflective surface, they transfer momentum \(2\hbar \omega/c) normal to the surface.  If the reflective surface is very thin and low mass, and you can bounce enough photons off it, you can get decent accelerations.  Part of the appeal is, this is a spacecraft where you effectively keep the engine (the whopping laser) here at home and don't have to carry it with you.  There are braking schemes so that you could try to slow the craft down when it reaches your favorite target system. A laser-powered lightsail (image from CalTech) Of course, actually doing this on a scale where it would be useful faces enormous engineering challenges (beyond building whopping lasers and operating them for years at a time with outstanding collimation and positioning).  Reflection won't be perfect, so there will be heating.  Ideally, you'd want a light sail that passively stabilizes itself in the center of the beam.  In this paper, the investigators implement a clever scheme to measure radiation forces, and they test ideas involving dielectric gratings etched into the sail to generate self-stabilization.   Definitely more fun to think about such futuristic ideas than to read the news. (An old favorite science fiction story of mine is "The Fourth Profession", by Larry Niven.  The imminent arrival of an alien ship at earth is heralded by the appearance of a bright point in the sky, whose emission turns out to be the highly blue-shifted, reflected spectrum of the sun, bouncing off an incoming alien light sail.  The aliens really need humanity to build them a launching laser to get to their next destination.)

9 hours ago 4 votes
Chatbot Software Begins to Face Fundamental Limitations

Recent results show that large language models struggle with compositional tasks, suggesting a hard limit to their abilities. The post Chatbot Software Begins to Face Fundamental Limitations first appeared on Quanta Magazine

2 days ago 3 votes
Links in Progress: We can still build beautifully

A tour of interesting developments built in the last two decades

2 days ago 3 votes
The Value of Foreign Diplomas

Is that immigrant high-skilled or do they just have a fancy degree?

2 days ago 9 votes
Incorruptible Skepticism

Everything, apparently, has a second life on TikTok. At least this keeps us skeptics busy – we have to redebunk everything we have debunked over the last century because it is popping up again on social media, confusing and misinforming another generation. This video is a great example – a short video discussing the “incorruptibility’ […] The post Incorruptible Skepticism first appeared on NeuroLogica Blog.

3 days ago 3 votes