While it's still fresh, pay attention to how it feels for a single training run to have just made the difference between "speculative risk" and "serious threat that the world is unprepared for."
And take a moment to be thankful that this particular threat is one that can be mitigated by choosing to not deploy it (and that the first ones to reach this capability are choosing to not deploy it).
If the industry succeeds at creating agentic machines that are capable enough to outsmart and outmaneuver humanity, choosing not to deploy won't be good enough. They'll just deploy themselves.
Need to run really large LLMs that require the power of RDMA with InfiniBand?
Check out this blog to see how to get it working on Azure Kubernetes Service:
https://t.co/IcvZyf8cST
#AKS#Kubernetes#InfiniBand#AI#LLM#RDMA
Ilya Sutskever at NeurIPS 2024 speaks about the forthcoming arrival of superintelligence: "that is obviously where this field is headed. This is obviously what is being built here" and superintelligent AI will have "radically different qualities and properties" than today's AI models
Robin Hanson says some time in the next century, AI will cause the economy to speed up from doubling every 15 years to doubling every month, resulting in most humans losing their jobs and we should insure against that scenario
Sequence modeling and design from molecular to genome scale with Evo
Researchers have introduced Evo, a groundbreaking genomic AI model designed to analyze and generate biological sequences at unprecedented scale and precision. With its advanced architecture and training on millions of prokaryotic and phage genomes, Evo represents a significant leap in understanding and engineering the genetic code across DNA, RNA, and proteins.
Evo’s ability to process biological data at multiple scales—from individual molecules to entire genomes—positions it as a transformative tool for decoding life’s complexity. It not only enhances functional understanding of genetic systems but also opens new frontiers in bioengineering, accelerating advances in genome editing and synthetic biology.
@dwarkesh_sp@ChrisPainterYup@gwern That digital avatar was a nice touch. Question: how did you enact Gwerns script, were there multiple teleprompters involved or just from memory? Nice job!
@leashless But the capabilities argument still holds, I think this is an argument about preferences, but we can train apparent preferences into a model. Certainly instruction following is very strong at this point.
Anyone who still believes that the development of AI has reached its limits should watch this video excerpt from Noam Brown (OpenAI).
Sometimes I get the feeling people arent listening closely.
Noam Brown: "We saw that like yeah it, once it's able to think for longer it um develops these abilities um almost like emergently that um were very powerful and contain things like backtracking and self-correction all these things that we were wondering how to enable in the models and to see it come from such a um a clean scalable approach um that was for me the big moment when I was like okay it's very clear that we can push this further and um and it's it's so it's so clear to see where things are going; no my think is understating how strong and effective his conviction in test time compute was."
Dario mentioned something in passing regarding the ability of humans to understand biology and what future AI systems might understand. The quote is: "there is lots of room at the top" around 13:11 reminiscent of Richard Feynman's famous quote
https://t.co/asu6Bh8r2X