maker of things and teller of stories
Waterloo, Canada
Joined October 2019
- Tweets3.4K
- Following1.6K
- Followers9K
- Likes6.7K
Come live with me!
highly recommend getting a 5ft bonsai tree for your room.
i’m also going on a sidequest and am subletting my 1bed in sf for a month starting next week oct 19.
lmk if interested! your roommates are @KrishRShah and @djkesu1
Use cases for this “sixth sense” ability are almost always latency constrained.
There are literally zero benchmarks that test this capability across reasoning levels and plot accuracy vs latency.
It’s a little sad how obvious this is yet nobody is evaluating the right things.
We’re introducing Humanity’s Sixth Sense, a new benchmark testing the intuitive visual reasoning people rely on every day.
Across 522 open-ended tasks spanning images and video, HSS tests everything from spatial and causal reasoning to social understanding.
The gap is significant. Humans score 93.1%, while the strongest model, GPT-6-astra, reaches 53.6%. The median model scores just 30.9%.
When shilling personal agents to normie friends, some respond with "I don't have much for it to do".
This is a limited framing. There is more to it than handling pre-existing clerical work.
They will make your live 10x better by doing the things you never had the time to do!
Well said, but I believe modern programming is more like "chiselling" than ever before.
A single prompt gives an oddly shaped rock of an implementation. Strike with another prompt and a chunk of stone pops off, getting closer to what you wanted but perhaps not there yet.
And you repeat this process, getting more granular overtime until you are happy with the outcome. Perhaps you don't know what is inside the rock, but at least you are pleased with the outside.
Krish Shah retweeted
"Sorry for the delayed response!
Middling generative AI has fundamentally altered the strategic balance between offense & defense in the war for our attention, so I have chosen to ignore this messaging channel.
Out beyond kino & slop, there is a field. I'll meet you there."
Would you rather be gpu rich but token poor OR gpu poor but token rich
53%Gpu rich, token poor
47%Gpu poor, token rich
62 votes • Final resultsUp until recently, it was enough to have some "proof of work" to prove that you are capable.
Nowadays you need to show "proof of struggle".
Have you allowed yourself to do the hard things? What is your pain tolerance when the agent stops working?
I’m back in Waterloo this weekend judging @HackTheNorth!
The ability to go from idea -> prototype has never been easier (much different than just one year ago).
Excited to see what people cook up, hit me up if you’re around :)
This is the first example of "we made a new algorithm" that seems legit.
Many use cases for modern LLMs involve asking a bunch of questions, and the prompt engineering involves figuring out how to ask the right questions.
After all this work, you are stuck with a slow + expensive system that you typically have to distill from.
Seems like a great solution, excited to try!
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
I miss the build in public days. At the time, there was not enough man hours to build everything so everyone enjoyed sharing the sauce.
Nowadays the idea holds much more value, hence a community that runs on secrecy and vagueposting.
AGIBOT releases GE-Act 2.0 — the first native World Action Model to validate a pretraining and scaling path for embodied AI.
📖 Explore the project: ge-act-v2.github.io/
Trained entirely from scratch on embodied manipulation data: visual representation, future generation, and action prediction, all from random initialization. No inherited video generators. No task-specific fine-tuning.
Put straight to a ruthless real-robot zero-shot test — unseen scenes, unseen objects, 100 atomic tasks, 20 skill categories, and two robot embodiments:
✅ Data scaled 100×: from 300 to 30,000 hours
✅ Task success climbs from 17.1% to 44.1% on G1-OP — with no sign of saturation
✅ New skills emerge at scale: folding towels, nesting paper cups, uncapping pens, arranging flowers
✅ Cross-embodiment transfer: G2-90D, under 2% of the training data, still gains 17.7 percentage points
✅ Failure data becomes a training asset — 2,000 hours of failed manipulations and deployment rollouts
A capable model envisions reality before it acts.
#AGIBOT #EmbodiedAI #WorldModel #PhysicalAI #Robotics
inb4 rappers talk about chargerless instead of APs and millies
We’re closing Chargerless orders soon bc too many people are buying it — if you want a health tracker that charges itself in sunlight you should order now on getchargerless.com
A sneak peek of production:
Every morning Claude asks me to choose from a bunch of high protein vegetarian dinner options.
Every evening—precisely when company door dash credits arrive—it orders for me.
This burrito was ordered with said system.
