Ben Heil retweeted
Our mentor Casey Brown, apart from being an extraordinary scientist, was a brilliant teacher who loved everything genetics. In his honor, some of the world's leading experts came together to create the 'Casey Brown Lecture Series' on human genetics 1/ shorturl.at/eCm7S
Ben Heil retweeted
So here's a story of, by far, the weirdest bug I've encountered in my CS career.
Along with @maciejwolczyk we've been training a neural network that learns how to play NetHack, an old roguelike game, that looks like in the screenshot. Recenlty, something unexpected happened.
Ben Heil retweeted
Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.
Ben Heil retweeted
In this spreadsheet I document a wide variety of h params on a per-model basis as well as provide my current recommendations for how to go about setting up a LLM architecture. Make sure to check out the notes tab for important caveats and warnings!
docs.google.com/spreadsheets…
Ben Heil retweeted
That coordinated quote retweet by OpenAI staff was the most impressive employee flex I’ve seen in tech. – at San Francisco, CA
Ben Heil retweeted
Very glad that our work with @GreeneScientist and others in integrating genetic studies with gene co-expression patterns is published in @NatureComms! Take a look at nature.com/articles/s41467-0…
Ben Heil retweeted
Our architectures get stolen by lesser AI teams,
So we better sign these rhymes with PGP;
But we'll keep on generatin' cash and tokens like a CFG
'Cause there's so much drama in the GPTs.
What's wrong, SaaS teams, am I making you nervous?
Even green VC teams are giving you denial of service.
You better not go broke before you IP;
Learn to prompt engineer, get friendly with the GPTs.
Framework control flow better than TCP,
The west coast's answer to Alan T,
Your model best not front if it can't pass the GRE,
'Cause there's so much drama in the GPTs. – at Stanford, CA
Ben Heil retweeted
It's time to celebrate Dr. Benjamin Heil! Ben was a member of the Casey Greene (@GreeneScientist) and John Holmes (@Johnhh429H) Labs, and now works as a machine learning scientist at Visa Research. 1/4
Ben Heil retweeted
Every day I see a thread about leaving academia to be happy and a subtweet saying that academia is great.
For computational students, this video by @rishiyer is great because it talks about different types of industry jobs and then compares in 12 axes youtu.be/H6wSxSXkr1w
Ben Heil retweeted
Science ran a big experiment on itself for the last ~60 years. It didn't work out.
experimentalhistory.substack…
Ben Heil retweeted
Ben Heil of the Casey Greene (@GreeneScientist) and John Holmes (@Johnhh429H) Labs defends tomorrow! #AlmostPhDone
ALT GCB Thesis Defense. Ben Heil. Neural Nets Are Not All You Need: Evaluating the Effects of Deep Learning on Transcriptomic Analysis. Advisors: Casey Greene and John Holmes. 1:00pm, Class of '62 Auditorium. December 15, 2022. Contact Anne-Cara Apple for the Zoom link.
Ben Heil retweeted
Eager mode was what made PyTorch successful. So why did we feel the need to depart from eager mode in PyTorch 2.0?
Answer: it's the damn hardware!
Let's tell a story about how the assumptions PyTorch were based off of became untrue, and why PyTorch needed to evolve. (1/10)
Ben Heil retweeted
Aran Komatsuzaki giving walkthroughs of the codeRL paper before the author arrives. After 10 minutes of SBFing his way into answering poster questions he revealed he was not the author and everyone lost their mind (Poster 138 #NeurIPS2022)
Ben Heil retweeted
1.0 release of Mamba (fast, drop-in replacement for Conda package manager) github.com/mamba-org/mamba/r…
Ben Heil retweeted
Came across this yesterday while at Rockefeller, should be required reading for anyone in a PhD or supervising PhDs! @CurrentBiology @betenoire1 Wish I had seen it sooner!
Ben Heil retweeted
Confused by all the "what is epigenetics" discourse?
Well I have just the thing for you -- presenting the Epigenetics Alignment Chart!
