ML interview question: why do embeddings come in 768 or 1024?
- “because BERT did it”
- “because of GPU optimization”
BUT WHY?!
The replies under this post is everything wrong with current courses and blog posts: superficiality.
this isn’t reasoning, it’s memorization
OpenAI fired this 23-year-old from their Superalignment team.
But he turned his insider knowledge into a $1.5B fund that's outperforming Wall Street by 700% this year.
He says maybe ~200 people in SF understand what's *actually* happening in AI right now.
Here's his thesis: 🧵
This is really cool. Irrespective of the field, AI simulations are expected to change the game. AI research is insane, and the pace of advancement in the field is much more insane. We may soon start to question the very nature of our own existence with our technology.
Simulations are the future, & one of the main tools we’ll ultimately use to understand and predict things about the universe. This is why I’m so excited about Genie 3, our latest interactive world simulator - here are some insanely cool things you might have missed about it 🧵:
@AbhnvSb @atullchaurasia I wouldn't say it's garbage. Although my journey wasn't in that particular order, I pretty much touched all he outlined. I am an active user of DL/ML for materials and drugs discovery.
@kmeanskaran@atullchaurasia Many useful tasks, especially scientific discovery, can still be done without MLops. For example, I have developed an ensemble of DL classifiers and trad ML models to identify anti-biofouling peptide sequences. All operations were from the stored binary files.
New paper! Fast matrix stress relaxation potentiates human monocyte 3D migration by generating protrusive forces and is dependent on the Cdc42-WASp axis. Thanks to our amazing collaborators! #viscoelasticity#mechanobiology@theChaudhurilab@Stanford_ChEMH
https://t.co/wR7HVrnITC
overseen in the recent GNN+ paper: gnn arch diagrams that are remarkably transformer-style.
this might help make gnns more approachable, and is appreciated. the results are cool too!
it further highlights nicely how gnns/transformers have compatible underlying symmetries :)
I am currently working on a self-moltivated project, actually an addendum to my previous research. The project, in summary, uses message passing neural network (MPNN) to classify small molecules as active or decoys. Why this? Think drug repurposing.
@HartungIngo@Glaconde34 Efficiently explained. Do you think this method will scale the challenges of acquired mutation that often render inhibitors ineffective?