Congrats, Kenneth! It was awesome teaming up with you to craft great developer experiences at Stripe. Can’t wait to see where your explorations at the intersection of LLMs and developer tools take you!
CODE
-
Finite-State Automata Features Generate Complex HTML Behaviors
By
–
Features connect in "finite-state automata"-like systems that implement complex behaviors. For example, we find features that work together to generate valid HTML. https://
transformer-circuits.pub/2023/monoseman
tic-features/index.html
… -
Monosemantic Features Steer Transformer Model Outputs
By
–
Artificially stimulating a feature steers the model's outputs in the expected way; turning on the DNA feature makes the model output DNA, turning on the Arabic script feature makes the model output Arabic script, etc. https://
transformer-circuits.pub/2023/monoseman
tic-features/index.html
… -
Second Cohort Completion: Chatbot Course Success
By
–
Me!! Just finished our second cohort: http://
chatbot-course.com -
Training AI Models Requires More Than Simple Code
By
–
Training a model of this type does not require merely writing 20 lines of code.
-
Building Software with AI and GPT-4 Vision Learning Paths
By
–
Ok who’s building the course on how to learn to build software with AI, GPT4 vision etc Got to be better paths than 100 days of python etc now right?
-
TPUs Integration Challenges with PyTorch Framework
By
–
TPUs are still a bit of a pain (to say the least!) to get working with PyTorch, and PyTorch is the easiest way still to get stuff done, on the whole.
-
Spatial Entity Embeddings: Historical Context Since 2017
By
–
The spatial locations thing has been around since at least the entity embeddings paper. IIRC that was around 2017 or so.
-
Spherical Deep Learning Outperforms Flat Space Models
By
–
Some well-rounded results: @GoogleResearch work shows that deep learning on a sphere — instead of flat space — is superior for things like prediction of weather & molecular properties.
— Jeff Dean (@JeffDean) 4 octobre 2023
Consider spherical surfaces (much better than pretending the world is flat!). See JAX code! https://t.co/8Q0msyJDcC pic.twitter.com/uGzb4piHYQSome well-rounded results: @GoogleResearch work shows that deep learning on a sphere — instead of flat space — is superior for things like prediction of weather & molecular properties. Consider spherical surfaces (much better than pretending the world is flat!). See JAX code!