Dario: Claude will take your job, but it will feel ashamed.
Elon: Look at this anime girl. She says the N word and is almost naked.
Zuck: ✨Superintelligence✨ will help people watch more instagram reels.
Demis: Gemini recently Calculated more precisely the motion of the Heavens.
Sam: Excited to announce that OpenAI Inc has entered into a definitive agreement with OpenAI LP and OAI Corporation, and has signed MOUs with OpenAI OpCo, LLC, and OpenAI Global, LLC!
It makes sense to pull the entire reasoning chain towards the hardware, but this means the programming model for LLMs is going to change. Instead of RAG injecting context from outside, I'm going to provide the LLM with lookup hooks.
This is a really good take. o1 makes sense to me now. You can think of it like doing work directly on the database with stored procedures so you can avoid the network back and forth.
the o1 model is great for compute cost performance, because it removes latency as the bottleneck (100ms vs 20s thinking time), which means more clusters around the world are viable at any given time to do inference, increasing liquidity
machine learning productivity is a mix between curiosity-driven exploration to understand every surprising observation and a focused pursuit of the otherwise singular goal of building a working system
as a longtime fan of Conway's "Game of Life" algorithm i never once considered visualizing the time axis physically -- this is breathtaking and elegant to me
credit:
https://t.co/qdPZiEAEPq
@swyx@wadefletch@DrJimFan https://t.co/76oM9NYOmu
^^ The release @DrJimFan is referring to. Light on the details, you can think of it as how GPT-4 isn't just regurgitating text, it's built up some kind of reality representation in its weights. The reality representation here is a physics engine.
@jxnlco You can't escape knowing your customer. AI is not a magic wand that waves away the need for specific context. But here's my question. This is all same-same, but different. What do you see as the "different" part?
@0xmaddie_ 2/2 And I would say time/locality-scale is very relevant when specifying dynamism. Do I really need to fundamentally rethink my value system every hour [insert shot at millennials here]? Not really. I can hold that as static and worry about more fine-grained stuff.
@0xmaddie_ 1/2 Could have both. The diagram provides context independent constraints (thanks for the book rec BTW, plowing thru it now), the structure for context dependent constraints to play out.
And now we're at this point where we're starting to recognize that even if you can't always pin down Truth, Goodness, Beauty, there's at least a structure and directionality to things.
The most exciting thing to study in 2024 is philosophy. We're at the cusp of this massive shift in thinking, to the point where there's alpha in sitting down and rethinking Platonic archetypes.
money, words, corporations, nations: all of these things are real, not "illusions". it seems important to describe the type of cognitive error that makes people think they aren't
Then we got postmodernism, which could be construed as more accurate, but with WAY WORSE user experience. It threw the baby out with the bathwater. Nothing was sacred anymore.
AI is in such a great honeymoon period, where the utility is still quite differentiated, and the oligopolists are fighting tooth and nail. It's just raining benefits down on consumers.