Semiconductor fabs are the cathedrals of our age
construction-physics.com/p/how-to-build…

I'm at #iclr24 this week with Guillaume Lample Devendra Chaplot @ ICLR 2024 Sophia Yang, Ph.D. @ ICLR 2024 Teven Le Scao & more from Mistral AI along with Rajko Radovanović @ ICLR 2024 Anjney Midha @ ICLR ‘24 from a16z ! come join us thurs night by signing up here partiful.com/e/YkvQX6VanlEv…

For those at ICLR, we are hosting a rooftop happy hour in central Vienna on Thursday evening together with our friends Mistral AI (cc Guillaume Lample Sophia Yang, Ph.D.), Robin Rombach and more in support of OSS AI!!

RSVP here - confirmations required:
partiful.com/e/YkvQX6VanlEv…

Contrary to claims SB 1047 would only impact AI megacorps, “covered models” include any non-derivative model that is as generally capable as circa-2024 frontier models. Algorithmic progress means in a matter of years, smaller players and even hobbyists *will* fall into its scope.

In 2023, hundreds of millions of $ were poured into lobbying regulators on ‘AI bio risks’ to justify stifling open source research

In 2024, common sense will fight back

Great piece by Data Innovation

datainnovation.org/2024/04/the-sh…

One of France’s smartest AI policies is their CIFRE program enabling PhD students to complete their thesis at industry labs with industrial compute budgets

Top frontier AI researchers return to France for PhDs after masters programs in the US instead of staying in the states

cooking

Tiny Mistral AI models everywhere

When I ask regulators this question, their true motivations and expertise are quickly revealed if their answer doesn’t start with “Why” AI-specific regulation is even needed in the first place

Anjney Midha Absolutely insane at this point. The AI genie is out of the bottle and any of this regulation only hinders growth and gives big business and foreign countries an advantage over us.

.Senator Scott Wiener what is the full list of startups that were consulted in the drafting of this bill?

something i learned recently from a turing award winner is that inventing neural networks 200 years ago does not qualify you as an expert at predicting the future and it definitely doesn’t make you good at crafting public policy

I finally read the full SB 1047 draft bill. Holy shit. This is serious Big Tech lobbying.

They should just rename it “How To Kill AI Startups and Open Source”

To make AI safe, you have to regulate its misuses. Not restrict fundamental research and open scientific progress.

.Garry Tan brings that Lee Kuan Yew energy San Francisco needs right now

Anjney Midha Tim Sweeney Tim was visited by ancient aliens and given the keys to the Akashic Record.

Anjney Midha Also, SideFx’s Houdini procedural modeling tool. There’s quite an art to enabling systems to be pipelined together.

Architecturally, generative models working in unison together look much more like a game engine runtime than an 'operating system'. Anyone building multi agent products today really shld be studying all of Tim Sweeney's blog posts circa '98

The most interesting work in LLMs right now is happening post-training. Teknium (e/λ) emozilla et all called it a year ago

Good news: bloated group, nothing will get done at the official meetings

Bad news: tons of backdoor meetings before/after the official meetings to push the real agenda

Worst news: this will 100% make AI less safe, less secure and more centralized

Can confirm Curio will make you an instant hit with nieces!