Before humans could write, how did we think? The leap from oral culture to literacy fundamentally reshaped human consciousness. What does that have to do with LLMs and AI today? My latest piece on substack:
hyperstellar.substack.com/p/magic-spel...
The new vision-language-action model can help robots operate in entirely new environments - clean a kitchen or tidy a bedroom that the robot has never seen before during training. A notable advance.
www.pi.website/blog/pi05
"So long and thanks for the fish" :-)
(in all seriousness, really excited for applications like this)
blog.google/technology/a...
Two cool real world applications of AI (Diffusion/DL) - 1. Designing antidotes for snake venom (>100K/year global snakebite deaths) and, even cooler, 2. Sifting thru venom-derived peptides to fight drug-resistant bacteria (snake venom has antimicrobial properties)
blog.asimov.com/p/antivenom-...
I think this is much of what great leaders do intuitively. By removing psychological roadblocks and introducing the right kinds of motivation and "arrogance", even, people can do impossibly hard things. Leadership has always been the art of responsively sculpting individual and group psychologies.
Or Arrogance or Self-Delusion for that matter: "I bragged about something so I’d have to perform... I have found that it paid to say, 'Oh yes, I’ll get the answer for you Tuesday', not having any idea how to do it... like a cornered rat I’m surprised how often I did a good job."
Courage, for example: "One of the characteristics of successful scientists is having courage. Once you get your courage up and believe that you can do important problems, then you can. If you think you can’t, almost surely you are not going to."
In which Gwern collects a long list of examples of how intelligence (as in solving very hard problems) is inextricably tied to seemingly unrelated psychological factors such as motivation, agency/empowerment, and perceived (not actual) difficulty.
“One of my fav Kaggle facts: after a long leaderboard stagnation period for a competition, seeing one team make a sudden breakthrough will often cause multiple independent teams to quickly reproduce the same breakthrough—with no knowledge of how the first team did it.”
gwern.net/on-really-tr...
"Imagine a fictional society that has unlimited wealth, unlimited health, and permanent peace. Would they be overflowing with joy? Probably not. I think their defining characteristic would be how trivial and absurd their grievances would be."
collabfund.com/blog/minimum...
Regardless of AGI & ASI, as soon as AI agents can own assets, we’ll have to deal with new ideas of person-hood. AI owning assets (crypto is a natural fit) will introduce a new kind of competitive surface with humans and new frontiers for what it means to be a person.
genesis-embodied-ai.github.io/videos/wukon...
This would've been mindbogglingly reeking of magic just a few years ago but now it barely makes the news (even if it's a still mindboggling).
Generative AI (VLM) leveraging Physics Solvers to create physically accurate simulations. AI + Physics!
genesis-embodied-ai.github.io
This. Narrative is sense-making. It is thinking. Descriptive writing processes sensory data through multiple lenses of meaning to create connections that may not pre-exist in the reader's mind.
HOT 🔥 fastest, most precise, and most capable hand control setup ever...
Less than $450 and fully open-source 🤯
by @huggingface, @therobotstudio, @NepYope
This tendon-driven technology will disrupt robotics! Retweet to accelerate its democratization 🚀
A thread 🧵
A whole bunch of systems that depend on effort being costly are going to be breaking.
Academic journals are seeing this happen already.
Dommaraju unseated China's Ding Liren to become the new world chess champion. "Every chess player wants to experience this moment, and very few get the chance," Dommaraju said after the match.
An illustrative timeline of how kernel smoothing ideas ultimately became "attention" in Transformers. This reminded me of Cosma Shalizi's earlier piece on this topic:
bactra.org/notebooks/nn...
Any space that AI can proliferate in will be flooded with slop, inauthenticity, and fraud (as is already happening)
Happiness doesn't always require positive emotional valence. Interview with Susan Cain - www.econtalk.org/susan-cain-o...
A thread of books I finished in 2024
Book 1
Lesson:
Nothing inhibits progress as much as the isolation between minds.
What do you do when you hit a plateau? Here are a few ideas:
hyperstellar.substack.com/p/a-guide-to...
In my latest essay, I explore Mastery by connecting the dots between eastern philosophical ideas, modern cognitive science, and the study of expert performers. Plus some unexpected analogies from training AI models. Enjoy!
hyperstellar.substack.com/p/how-to-mas...
John McLaughlin at 81 still has the chops, and so does Zakir at 72. A wonderful 20 mins of Shakti performing at Tiny Desk.
www.youtube.com/watch?v=Cx8A...
Nice framing of "unwanted drives" in AI without invoking agency
"... if a model hallucinates a fact, it is unwanted behavior (but not drive); if it insists on the hallucinated fact & works to convince the user it is true... that would be an unwanted drive."
bounded-regret.ghost.io/intrinsic-dr...
"So if ultimate impact on the world is what you crave, becoming a top science fiction writer is probably the best bang for your buck."
Love this framing as a reason to write sci-fi:
www.strangeloopcanon.com/p/so-you-wan...
A long-form essay on what AI can reveal to us about human cognition. In which I somehow connect Elena Ferrante with LLMs. Enjoy!
hyperstellar.substack.com/p/let-me-fin...
