Huh TLT is lower than it's been for 20 years I wonder why that is
Not enough people are outright saying it: AI doing well means interest rates go up. All money being lent is going to go to data centers, because they're lucrative. This means rates go up (unless Fed allows inflation). No way out of that.
Preregistering my belief that LLMs will cross human capability levels and then design the next architecture that will be more bitter lesson pilled.
Same pattern as AlphaGo -> AlphaZero
LeCun will then be right on a technicality
"LLMs arent bitter lessoned enough!!"
You know what else wasn't bitter lessoned enough? AlphaGo.
LLMs will surpass humans before needing to be upgraded. Then LLMs will build the new Bitter Lesson architecture to surpass themselves.
Fundamentally this is the same argument as "scaling llms is dumb let's just make specialized ai for every single use". It will always be both and thinking specialization "ultimately wins" is just naive.
.@mcuban says humanoid robots won't last more than 5-10 years.
Instead, we'll "design the house to fit the robot, and design the robot to fit the house."
"You could create a house where the pantry, the refrigerator, and the washing machines were hidden behind the garage, if you even have a garage. That way you could redesign the house so that all the living space was for people."
From his appearance on the show in March.
Bad Copy: Datacenters use less water than agriculture
(but we need to eat!!!)
Good Copy: Datacenters use less water than golf courses
Sam Altman keeps saying AGI will create lots of new jobs. I asked his own model and every other frontier model. They all disagree with him. If AGI is real and cheap, then economically valuable jobs go away.
People going crazy about datacenter water usage are indeed wrong. However one could argue they are "directionally right" in the sense that govt. really is doing nothing about the impending automation wave, and that's bad.
Realizing that just watching Polymarket for Claude and GPT 5.5 release dates is way more useful than anything the AI leakers have to say tbh.
I joined Google's AGI benchmark competition on Kaggle. Very excited, I spent quite a few days creating three new stratified, unsaturated benchmarks.
Turns out, you can only submit one per person. Oh well 🫠
I swear if these people aren't heavy shorting NVDA and just blowing hot air instead I am going to CRASH OUT
"Tapestry woven from threads"... This basin has existed for some time...
"Not metric driven" hardly means anything. Literally set the metric as "Novel, interesting, testable idea in x domain"
Ta da! Your field is now hill climbable by Claude.
GPT released two new models and I ran them on TopoBench (Lean-verified Topology). Nano beats Flash-lite at same price. Kimi and Deepseek beat Mini handily at a lower price. Best cheap American models, not best globally for math proving.
So many anti-AI heads need to be asking themselves a simple Q. Are models improving, Or not?
i am using supermaven again and i have something to say about this whole AI thing.
I think as a group (swe) we rushed so fast into Agents when inline autocomplete + actual skills is crazy. A good autocomplete that is fast like supermaven actually makes marked proficiency gains, while saving me from cognitive debt that comes from agents.
With agents you reach a point where you must fully rely on their output and your grip on the codebase slips. Its insane how good cursor Tab is. Seriously, I think we had something that genuinely makes improvement to ones code ability (if you have it). Truly acts as a multiplier, and we left it in the dust because it is not sexy.
hurts me on the inside.