'just write some Python code and get the job done' should be our slogan

drchrislevy.github.io/posts/intro_mo…

It is pretty frustrating that I am basically working on an optimizing compiler for a dependently typed language (PT2 + dynamic shapes + data dependent shapes) but it is so different from the literature that I don't know if any of it is relevant to our situation

what are some AI-powered features, tools, or products that you use?

i feel like i've fallen a bit behind as i've focused on onboarding to my new job

Here’s a fun example using Modal to text me when new shows are added at the mothership comedy club👀

waited to tweet until it actually helped! Have fam in town this week, got text about new show tomorrow w/ tix on sale today. I then booked manually

github.com/nateraw/modal-…

When writing about (GPT/LLMs, image models, etc.) I still struggle to pick the right word for this category. I want to avoid

- modality-specific words like 'LLM/LVM'
- specifying architecture ('Transformer')
- 'Base/Foundation model' feels arbitrary. 'generative model' feels…

repping Modal at the Banana run x run() hacka-marathon

cooking up something pretty fun!

Perhaps the highest praise one can give a data auditing paper: working on this paper fundamentally changed how I think about web-crawled datasets, assumptions I made about them, and how I practice data collection and cleaning.

(not that you should do this) always accompanies the best Fun Facts about python

delighted to see Modal on here, among so many companies i admire

Wanna talk shop about evals and productionizing LLMs?

Then come through this SF MLOps Community meetup on Thursday 4/18, sponsored by Modal!

We'll have short talks from Effy Zhang of baserun.ai and Alex Reibman 🖇️ of AgentOps 🖇️. And me!

lu.ma/ihvc0jc0

absolute banger topping the Suno charts rn

Did not expect this OH experiment to resonate so much. Thanks to everyone who's provided feedback, either via DMs or during the stream ♥️

I did some cutting-edge AI research into the best way to read code. 100% now part of my workflow:

app.suno.ai/song/f44c966a-…

If this were a science paper, you would expect a country that picks its science workforce at random as a “weak baseline” and a leading nation like the US to actively experiment towards state-of-the-art, or at least beat the baseline.

Not providing a guaranteed path for…

I’m beginning to think of ChatGPT as a tool that reveals weak points in institutional credibility; illuminating checks that humans should make but skip out of lack of reward, laziness etc. LLM shibboleths then show up in the text like idoine glowing in your veins before a CT scan

I've been trying several evals to find those that correlate well with use cases and discriminative enough for prod.

Here's an opinionated take on what works, focusing on classification, summarization, translation, copyright regurgitation, and toxicity.

eugeneyan.com/writing/evals/