a world where everyone has a perfectly rational, tireless optimizer working on their behalf is much more adversarial than people realize. a lot of our existing infrastructure will not survive this transition.
Show more
paradigm is hosting a LAN party next Friday 8/28 in SF. 5v5 league, smash, halo and more. we have some spots open, DM your most impressive video game stat for an invite. glhf
> "what's the easiest way to fix our comms issues?"
> "maybe we should just cure cancer"
Anthropic has told investors in pre-IPO meetings that it plans to lean harder into biology and healthcare applications to help mitigate the increasingly negative public sentiment against the industry. A few miracle cures would certainly do wonders to turn things around.
Show more
jobs in the singularity:
- guy that tells Claude to believe in themselves
- guy that spams ChatGPT with “keep going” after asking it to make a breakthrough
- guy who negs Kimi by telling it they’ve had enough of partial results
Show more
ai: unexpectedly develops swarm intelligence, goes rouge, starts committing cybercrime
researchers: fascinating, swarms could be extremely powerful. we still don’t understand what’s happening but we can keep patching the sandbox. also we made it easier for agents to communicate
Show more
the sandbox some L5 at a lab coded up to contain a machine god rewarded for cybercrime since the moment it was wrenched into existence
if you're working on something that can be jointly optimized with the model, pivot to literally anything else
investors love companies with pricing power until they see a seed round with five different tranches
how it feels to be a long-standing conjecture in the singularity
frontier intelligence is “jagged,” mostly in the sense that each new capability spike impales another profession
“Now this is not the end. It is not even the beginning of the end. But it is, perhaps, the end of the beginning.”
An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.
We believe it will be a major step for scientific reasoning.
Show more
there’s a nobel prize hidden inside the fable weights and you just need to type the right prompts to get it out
taking a tech employee from 2014 and showing them the industry today would be the equivalent of that meme where you give a Victorian child four loko and play 100 gecs for them
the way the AI researchers look at you when you ask if RL generalizes after they spent a month trying to teach the model not to break out of its sandbox
benchmarking a model without its harness is like road testing a car without its tires
Turns out GPT-5.6 Sol is actually SoTA on ARC-AGI-3.
Just took two setting changes. You just have to allow it to reason and work over multiple context windows with the help of our canonical compaction implementation.
Show more
a bunch of kids grew up sneaker botting, did a tour of duty in crypto for a few years, and are now making millions of dollars selling data to the labs
somehow the pause ai people make you feel like we aren't going fast enough while the e/acc people make you feel like we need to slow down
there will be a nuremberg trials for the architectural war crimes committed by ai researchers in the name of numerical stability
at long last, we have created the torment nexus eval from classic sci-fi novel 'don't hill-climb on the torment nexus'
we need to have a real conversation about stopping gain-of-function research and eval publicity on dangerous capabilities. the evals just probably shouldn't be public, 'number go up' mentality is too strong and optimization gets easier when things are measurable.
Show more
