But, like. There is a pretty good chance we get some kind of Star Trek utopia. And the fact that an equally likely outcome is some unassailable system of control that capitalism or another authority structure wields against humanity until we're toast just isn't enough of a demotivator for the optimists ¯\_(ツ)_/¯
Well, there are a lot of scientists [not tech bros, though there are many of them spamming nonsense left and right] who are of the view that faster == better.
There is a huge spectrum of perspectives on risk, but the smallest minority are people who see no risk at all. The more common one is the realpolitik of an "AI space race" will probably destroy us skynet style. Already happening with DoD hastening their timeline to deployment of autonomously lethal systems.
lol are you for real XD
realtime life or death decision making has very little to do with visual effects. human vision is [barely] sufficient to the task of driving. it's [still] silly to imagine a system with less sensory data is going to be as/more reliable than one with more.
@evidence_dev Is there any chance y'all could look at @prql_lang as a stand-in for standard SQL syntax? I have long thought it's a perfect fit for this project.
Also wondering if you've considered finetuning an LLM on multiple examples and/or offering a "GPT" that speaks Evidence.
Cashflow positive isn't the same as product-market-fit. The latter implies a takeoff where your active customer's demands determine the shape of your product's roadmap, doesn't it? I think companies have experienced positive cashflows without the gravitational pull of PMF, at least so I've read
Nobody's heard of it because @AMCPlus memory-holed it for TAX purposes even though Season 2 was already in the can!!! Grrr.... Anyway thanks to Amazon NZ/AU and the wonderful world of piracy, you can finally watch it. HMU if you need access. Scavenger's Reign is a tasty morsel but Pantheon is a whole, satisfying meal
This has been a long time coming. I'm not sure if you're familiar with Ycombinator but they're a startup seed fund who have issued "request for startups" in the past, including the one from 2012 called "Kill Hollywood" [which they've since deleted but which is accessible from the wayback machine]
https://t.co/0B5FiRWRW3
The short term impact may be felt primarily by crews, casting agents, and caterers... but long term the idea of new movie stars becomes kind of hard to imagine.
The deal makes sense for those who are already famous, as they can potentially make more $ for merely signing off on work. Everybody else will be worse off. This pattern will repeat in other industries.
How do you mean, status quo?
I think it's true that we're arming them and it's because a conflict is likely unless [perhaps even if] we do.
No matter the cost or current rhetoric, the west will defend Taiwan even ahead of Israel and Ukraine. AI > * when it comes to risk. TSMC, and the world's reliance on it, is Taiwan's true shield.
Can GPT-4 teach a robot hand to do pen spinning tricks better than you do?
I'm excited to announce Eureka, an open-ended agent that designs reward functions for robot dexterity at super-human level. It’s like Voyager in the space of a physics simulator API!
Eureka bridges the gap between high-level reasoning (coding) and low-level motor control. It is a “hybrid-gradient architecture”: a black box, inference-only LLM instructs a white box, learnable neural network. The outer loop runs GPT-4 to refine the reward function (gradient-free), while the inner loop runs reinforcement learning to train a robot controller (gradient-based).
We are able to scale up Eureka thanks to IsaacGym, a GPU-accelerated physics simulator that speeds up reality by 1000x. On a benchmark suite of 29 tasks across 10 robots, Eureka rewards outperform expert human-written ones on 83% of the tasks by 52% improvement margin on average. We are surprised that Eureka is able to learn pen spinning tricks, which are very difficult even for CGI artists to animate frame by frame!
Eureka also enables a new form of in-context RLHF, which is able to incorporate a human operator’s feedback in natural language to steer and align the reward functions. It can serve as a powerful co-pilot for robot engineers to design sophisticated motor behaviors.
As usual, we open-source everything! Welcome you all to check out our video gallery and try the codebase today: https://t.co/BHiNmqPoWE
Paper: https://t.co/bdh9TYQtHm
Code: https://t.co/lqKiaM2yYJ
Deep dive with me: 🧵
@julien_c@AdeptAILabs Would you really? I'm not sure it's sound business to rely on a small free model from the people who have the massively superior big boy hidden away behind an $api [presumably, eventually. or their product could be a service with no api, but I doubt that in the long run].