ML @ Meta, past: ML @ Twitter | 🎸🏃‍♂️🚴‍♂️

New York, NY
Joined November 2011
sending your way a bit of snow & fresh air
3
1
1
38
Plato thought writing would weaken the minds. Baudelaire considered photography as corrupting art. Sousa that recorded music would kill amateurs. Tao that AI advancing science is not desirable. Carry on
Replying to @KrzakalaF
Au contraire. C'est une nouvelle ère qui s'ouvre pour les mathématiques. Une ère où la démonstration formelle est largement automatisée et où l'accent sera reporté sur le développement de nouveaux concepts, nouvelles abstractions, nouvelles définitions, et nouvelles conjectures. L'invention du bateau a réduit l'importance de la nage, mais a permis la découverte de nouvelles terres.
1
87
The best Christmas presents are the ones you buy for yourself 🎁 @LeRobotHF
2
358
Claude decided to delete my database ☠️ cc @bcherny @trq212
1
3
395
Coding is dead, long live coding
1
1
2
411
The expectations for OpenAI's new products have always been high - I find GPT-4o extremely impressive, getting to that quality (LMSYS elo of ~1310 so far, better than any model) with that speed is crazy good in my book. The voice mode is the 🍒 on the cake - a very big one
1
1
398
I can't help but wonder how they can offer that level of compute though - "always on" audio + vision models feel expensive on paper. They might have 2 models, a cheaper one in front of the expensive one to filter out inputs "not worthy" of compute.
256
. @dwarkesh_sp’s podcasts are great - consistently bringing interesting guests to discuss tech and some adjacent topics! Thank you @dwarkesh_sp
Zuck on: - Llama 3 - open sourcing towards AGI - custom silicon, synthetic data, & energy constraints on scaling - Caeser Augustus, intelligence explosion, bioweapons, $10b models, & much more Enjoy! Links below
3
439
Is there a good application that allows to use an LLM directly from the right click menu on Mac? You predefine 3 or 4 prompts and you can summon the LLM on the fly with any application through a right click.
1
214
This is an interesting take on an industry shift, however I think the changes will be more drastic! Here is a potential future, somewhat of a thread.
We're going through what is likely a major industry shift - and it will change how many (even though not all!) engineering teams think about what "sensible" engineering practices mean. Say goodbye to this logic from the past: 1. Go 2. Microservices 3. ??? 4. Profit
1
554
End to end tests will become more common as code become more ephemeral and LLM make it easier to write them.
1
1
223
Due to these factors, outages will continue to happen more often than before, however resolution will be faster in average.
163
Ok I tried suno.ai V3 alpha, it is very impressive. Way better than their v2 model, instrumentation is too notch, a lot of the styles are pretty good. Voice is not fully there for the most part (seems "autotune") but it sometimes can sound still quite realistic
1
7
475
A song about not very good at surfing but still trying (soft / acoustic pop rock) app.suno.ai/song/ac0c2e93-1e…
78
Mark, want to rename Threads too? I heard Elon has budget issues and does not need the Twitter brand anymore… maybe, just maybe.
3
217
This is v cool. Local LLMs feels like the future due to current costs / privacy concerns. Yet we saw what happened in the past 10 years to private solutions for OS / mail / storage / messaging options. Avg consumer buys the cloud version. (1/2)
I “jailbroke” a Google Nest Mini so that you can run your own LLM’s, agents and voice models. Here’s a demo using it to manage all my messages (with help from @onbeeper) 🔊 on, and wait for surprise guest! I thought hard about how to best tackle this and why, see 🧵
2
5
762
I wonder if labeling would help. Think "cloud" versus "local", similar to organic versus non organic labeling exists today. (2/2)
128