Let’s create life. CTO at @comma_ai
Joined November 2022
- Tweets1.1K
- Following253
- Followers7.6K
- Likes6.6K
Pinned Tweet
The world is not zero sum. People can create something out of nothing, and many do. Should be encouraged more.
openpilot 0.11.2! The new big driving models have ~100x more FLOPS. This is the start of a new era of intelligence for openpilot.
blog.comma.ai/0112release/
Looking for ideas on RL training recipes.
We have 10M miles of driving data, mostly human.
We want to make a policy that drives as well as a human, or better. Assume we need to train with RL inside the simulator below.
What's a good reward and training recipe?
4 more years? yassineyousfi.github.io/page…
Harald Schäfer retweeted
4 more years? yassineyousfi.github.io/page…
I'm not worried AI will end humanity, and you shouldn't be either.
But I am worried it's going to make the future lame. History has been wonderfully exciting, and I want the future to be equally so.
I don't care about your P(doom), but let's make sure your P(lame) is low.
You can install openpilot on a 2019 corolla, an 8 year old car, and it will stop at stoplights for you.
If @comma_ai didn't exist, old cars would almost certainly never get capability like this. And most people would claim a project like openpilot could never exist.
Many cool things only exist because a few people decided it should. You can just do things, and more people should!
The new location will not be in California, because power is criminally expensive here.
We'll need robots to remotely fix things sometimes, also hiring people to make those robots work!
Our GPUs have outgrown our little San Diego office. In 2027 we're starting up a bigger better datacenter with more GPUs. We are hiring engineers to work on this.
If you know linux and like sshing into machines that cost more than a Ferrari, you might be a good fit.
More info on cluster: blog.comma.ai/datacenter/
Hiring both interns and FT engineers. FT salary 100-200k. My email: [email protected]
I've always been an AGI/foom skeptic, but I admit modern LLMs feel a lot smarter than I expected they would. It's hard to find things they cannot do.
Yet all real-world value metrics (GPD, life expectancy, production, ..) are stable and indicate this is just technological progress as usual.
I find it hard to reconcile those two things. My best guess is that we just don't understand intelligence and value-creation. And that our primitive intuitions about them are easily fooled, but time will tell.
openpilot's longitudinal policy has 2 modes. Traditional ACC, and full e2e mode where the driving model decides everything. Including highway speeds, slowing down for curves, and stopping at lights.
Since this year users are preferring the e2e mode! Time to go out of beta.
This is incredible news. Big companies having monopolies on cyber capabilities would be terrible.
The future is much brighter when power is decentralized.
Robotics and self-driving will go this way too. It will take a while, but open-source will eventually dominate.
Hopefully openpilot will speed this along!
Nice work! Cool to see people use comma 4s for this type of stuff, the power of open-source!
