vishal

Claude can actually do CAD now in @Onshape Here it worked for an hour and built a 4-part monitor arm, starting only from a sketch and description. The trick was to give it the tools to look at its own work. Introducing: Jarvis Onshape MCP

NASA reports signs of past life on Mars... youtube.com/shorts/p5nvGbk…

never seen before robotic agility...

boom. supersonic commercial aircrafts are back. Boom - XB-1 (boomsupersonic.com)

the bottom line (in the red box). arvind @random_walker's straightforward approach to exploring AI's social and economic impact is both refreshing and deeply insightful.

Congratulations, India🇮🇳! A proud day for every Indian. Jai Hind!

We are in the #HelloWorld era of transformer based AI. I wonder what the #Kubernetes era of this AI will look like.

@paulg @benparr I think you meant transformer-flavor AI.

AI is probably as big a deal as microprocessors or the web. But for the first few years only nerdy enthusiasts cared about those, whereas everyone is using AI.

@gmiller @paulg This sounds off by a factor of 9. 11 days ago @sama said the number is around 150 m at ASU+GSV Summit m.youtube.com/watch?v=-PUApY… (also this article looks like a GPT piece. Convincingly telling you a lie.)

@paulg It is striking: apparently over 1 billion people have tried Chat GPT, according to this site demandsage.com/chatgpt-statis…

“Today’s ‘AI’ revolution is more of a UX revolution than an AI revolution.” — @quaesita link.medium.com/FaYl6bwllzb

@tunguz It's very likely. Words like "Black Box" and "AGI" have confused people a little bit. People don't know how weak and incapable these models are.

Getting a sinking feeling that most of the AI doomers never actually trained a single ML model in their lives.

@ylecun Blockchain certainly comes to mind but figuring out who own what is very convoluted 🙂

In the age of computer-assisted creation, how do we seamlessly propagate rights to remix and backpropagate credit (and money)?

I can attest to that intuitively as someone who has done really well on GRE and not so well on JEE… 😂

Sensing a calm before the storm. AI breakthroughs have been normalized now. There are a few dozen algorithmic innovations in the text, image, 3d, and video AI space every week. It's time to learn & build.

There is a middle ground in this debate.

Now an average joe has gotten the super powers. LLMs are certainly not capable of automatically doing things the Terminator-style, so no worries there, but the products that can reach masses might need regulations (as you also suggest in your interview with @AndrewYNg )...

@ylecun I think in your quest to defend your point on AI safety you are pushing too far on one side here. LLMs are far more powerful than Google answers. Plus people do have some legitimate concerns related to the power of AI getting into wrong hands...

This week on Construction Physics - how did solar power get cheap? constructionphysics.substack.com/p/how-did-sola…

@DrJimFan I think you are talking about emergence in LLMs here. @_jasonwei and his colleagues explain is well in this paper: Emergent Abilities of Large Language Models openreview.net/forum?id=yzkSU…

Many of you ask, why can LLM ever be superhuman if it’s trained on human data? The reason is simple. It’s far easier to judge whether a work is good or not, than to create the work yourself. You can be a food connoisseur but terrible at cooking. You can be a top art critic but terrible at painting. Feedback signals from a legion of expert humans will eventually produce an AI stronger than all of them combined.