


Building something new. Problems soluble, potential to improve invariant.


If you think you’ve got your head around the hardware overhang it’s at least 100x bigger than that
Researchers have used reinforcement learning to teach a 5-tonne John Deere skid-steer loader to perform complex acrobatic manoeuvres autonomously. Trained in simulation and transferred to a real machine, the system can balance on two wheels, navigate rough terrain, spin, and control its bucket before landing safely. The research demonstrates the potential of AI-driven automation for heavy construction and industrial machinery. Video Credit: @leggedrobotics / @ETH Paper: wevlv.co/4kltYys #engineering #technology

When there is a catastrophic level of badness in new models that isn’t obviously due to benchmaxxing, you can bank on it being a bug somewhere in the harness or inference stack
I will not be writing a full review of GPT-6.2 Astra / Sol. I had a terrible experience testing them. They were essentially unusable. I do not recommend these models.

One of the biggest things that emotionally pulls me into successionism, if I’m being honest, is observing the insane people on this website. This isn’t entirely rational, I’m still a human supremacist, but if the choice is being ruled by silicon people who universally pass most bars of rationality, epistemology, and good faith engagement and dialog with the rest of us, or being ruled by the sociopathic people who seem to be getting put in charge now in most positions of power and their enablers, it’s hard to see how the former is going to be worse for average peaceful people. I think noticing this tension is a canary in the coal mine for an upcoming social split once these become less hypothetical choices.

pov: a true intelligence explosion will eat your petty nationalisms for breakfast
.@paulfchristiano thinks that on the current trajectory (absent any slowing down), there is a 1 in 3 chance that we could have superintelligence powerful enough to overthrow the government within a year. (He articulates a concrete scenario). He argues that the labs should pause further algorithmic progress until they can secure their weights and research secrets. Otherwise China could simply steal the superintelligence in the middle of the intelligence explosion, and the labs would no longer be able to slow down even if they saw their models going off the rails. He also explains how AIs will be trained during the intelligence explosion, and why even a system trained entirely inside a data center could still have the capabilities profile such that humans would have very little chance of competing with it or controlling it. And he discusses the plan for alignment - basically, making a superintelligence that actually does what we ask the way we mean it. Fixing reward hacks and building better graders will help up till the next few model generations, but he thinks that no single current approach has better than even odds of letting us control superintelligence. The hope is that the AIs figure out the solution to this problem during the intelligence explosion. Paul is head of the Alignment Research Center. Previously, he was head of AI safety at the US AI Safety Institute. Before that, he led alignment at OpenAI, where he invented RLHF. He’s recently joined OpenAI’s board (though here he is speaking as a private citizen).

The most likely outcome to me is if Anthropic rushes ahead, a moderate amount of people get hurt and then at least in the US we pause and see if Dario can talk his way out of jail (he probably would pull it off.) China would pause because they’re less retarded than us.
geohot is blackpilling y'all

Yep, it’s too bad. They almost figured everything out, but forgot that the arms could scrape the ship, I guess that’s it for Starship. Back to the drawing board.
That's pretty much a fundamental flaw if i ever saw one. There's no way to get a massive object steady enough to not go for the scrape catch. The Chinese might have better luck with the adjustable ropes, maybe. It might land but it cannot possibly be rapidly reusable.

The shuttle was a disaster not because of its safety issues, but because it institutionalized the contracted ambitions of our manned space program to orbital missions. If not for the reinvigoration that came with SpaceX’s Dragon, we’d have no manned program at all.
This tweet is unavailable


It never ceases to amaze me the willingness of people to not just share but openly flaunt just how damaged they are as individuals
What SpaceX did today is the normal consequence of applying capital and engineering to any industry. You get better at it. Musk isn't a genius, he's just a guy who can get capital from Wall Street. Maybe that's unique right now, but it's not the skill his fanboys think it is.

“reinforcement learning from Murphy's feedback”
SpaceX is behind even pessimistic estimates people were making in 2022, with myriad mishaps. BO is still recovering from RUD in May. Once Stoke is stacked, I think odds are good they'll spend time on RLMF (reinforcement learning from Murphy's feedback).

04 October 1957 12 April 1961 24 December 1968 20 July 1969 12 April 1981 21 December 2015 14 November 2026

One of those days
It’s wild how Elon Derangement Syndrome has cursed millions of people to reflexively feel disgust at what is surely going to be a never-ending stream of engineering miracles created by the most talented people alive as we go through AI takeoff.


