Links for 2023-02-25
OpenAI’s planning for AGI: “…we are going to operate as if these risks are existential…A misaligned superintelligent AGI could cause grievous harm to the world; an autocratic regime with a decisive superintelligence lead could do that too...It’s possible that AGI capable enough to accelerate its own progress could cause major changes to happen surprisingly quickly...Many of us think the safest quadrant in this two-by-two matrix is short timelines and slow takeoff speeds; shorter timelines seem more amenable to coordination and more likely to lead to a slower takeoff due to less of a compute overhang, and a slower takeoff gives us more time to figure out empirically how to solve the safety problem and how to adapt.” https://openai.com/blog/planning-for-agi-and-beyond/
Impressive: Meta AI has released LLaMA, a large language model with only 65B parameters that achieves results competitive with 540B parameter models such as PaLM using publicly available datasets. It even performs well in mathematics, although it is quite below Minerva-62B, which was fine-tuned on mathematical datasets. “All our models were trained on at least 1T tokens, much more than what is typically used at this scale. Interestingly, even after 1T tokens the 7B model was still improving.” https://threadreaderapp.com/thread/1629151231800115202.html
“How to teach robots to perform long-horizon tasks efficiently and robustly🦾? Introducing MimicPlay - an imitation learning algorithm that uses "cheap human play data". Our approach unlocks both real-time planning through raw perception and strong robustness to disturbances! MimicPlay is a hierarchical imitation learning algorithm that leverages cheap and non-labeled human play data (10 minutes) for learning the high-level planner and a small amount of robot data (20 demonstrations ~ 20 minutes) for learning a plan-guided low-level controller.” https://mimic-play.github.io/
Using AI tools to design an entire landing page: ChatGPT for text+visual design+Javascript, ElevenLabs voice + D-ID video, Midjourney+Photo Room art/photos https://www.samdickie.me/writing/experiment-1-creating-a-landing-page-using-ai-tools-no-code
China tells big tech companies not to offer ChatGPT services — State media outlet blasts chatbot as spreading U.S. government 'misinformation' https://asia.nikkei.com/Business/China-tech/China-tells-big-tech-companies-not-to-offer-ChatGPT-services
CRISPR pioneer Feng Zhang launches new genetic delivery startup with $193 million https://www.statnews.com/2023/02/16/feng-zhang-aera/
3D-printed organs may soon be a reality. ‘Looking ahead, we’ll not need donor hearts’ https://fortune.com/well/2023/02/15/3d-printed-organs-may-soon-be-a-reality/
“After watching how people use ChatGPT, and ChatGPT's weaknesses due to not using inner-monologue, I think I can be more concrete than pointing to non-robust features & CycleGAN (or the S1 'blob') about why you should expect RLHF to put pressure towards developing steganographic encoding as a way to bring idle compute to bear on maximizing its reward. And further, this represents a tragedy of the commons where anyone failing to suppress steganographic encoding may screw it up for everyone else.” https://www.lesswrong.com/posts/bwyKCQD7PFWKhELMr/by-default-gpts-think-in-plain-sight?commentId=zfzHshctWZYo8JkLe
Marc Lackenby - Using machine learning to formulate mathematical conjectures https://www.youtube.com/watch?v=0ekP5M7w3dQ
“Scorpius X-1 puts out about 23000000000000000000000000000000 watts of power in X-rays! Yes, that's 2.3 × 10³¹ watts. This is 60,000 times the X-ray power of our Sun.” https://mathstodon.xyz/@johncarlosbaez/109875481393123825
A pulsar was disrupted, producing a kind of antimatter spike extending out 7 light years or so https://arxiv.org/abs/2202.03506
Astrophysicists Discover a Mysterious Perfect Explosion in Space – “It Makes No Sense” https://scitechdaily.com/astrophysicists-discover-a-mysterious-perfect-explosion-in-space-it-makes-no-sense/
Chinese astronomers have found evidence of a "dark" galaxy: 98% dark matter, 2% hydrogen, and almost no stars at all. It showed up in their radio telescope, but when they looked for the light from its stars they saw...nothing. https://www.sciencealert.com/mysterious-dark-galaxy-emits-no-visible-light-scientists-say
Huge study involving over 7 million individuals finds that the ingroup bias (preference for one's own group) of black Americans is approximately 5 times stronger than the ingroup bias of whites. https://pubmed.ncbi.nlm.nih.gov/35895290/
Deaths from sepsis following the infection of a minor cut or blister were once extremely common and no amount of wealth or power could save a patient. https://www.humanprogress.org/u-s-presidents-son-dies-of-an-infected-blister/
😮🚀
I've always been an avid reader of science fiction, but it's lost a lot of its appeal now that we're on the eve of the singularity. Instead, I just digest the latest AI rumors and breakthroughs.
Anyone who needs to talk AGI or AGI strategy needs to understand this table from Drexler's _Nanosystems_, his PhD thesis accepted by MIT which won technical awards. These are calculation outputs; you need to have seen how Drexler calculated them.
— Eliezer Yudkowsky