AI Reading for Saturday Jan 4
Can LLMs Make Robots Smarter? - ACM
“Language is not related to the hard problems of robotics in any way.” says one senior professor.
Fast-learning robots: 10 Breakthrough Technologies 2025 - MIT Technology Review
Up close with Amazon's Sparrow picking robot - YouTube
o1 cheats at chess - Geeky Gadgets
Rogue not rouge dammit. LLMs are fundamentally imitative, skin deep AI. I think when o3 launches it will be something like
Solves > 50% of Putnam math problems because it has seen them before
When you fuzz them it solves 1/3 of them but partly because it still recognizes some of them
It solves 25% of novel problems by throwing spaghetti at the wall, tries something and gets it mostly correct, sends it to a critic and is able to fix it some of the time. Meanwhile on 50% of them it goes totally off the rails.
No doubt a process like that is incredibly helpful and productivity force multiplier in many contexts. But you can’t trust it for high stakes autonomous activity without some breakthrough in knowledge representation and a proper world model. Not really AGI until it learns on the fly and adjusts in game theoretical optimal or exploitative ways, b/c otherwise it is always susceptible to adversarial shenanigans.
If you tell AI what group your are in, it answers with in-group bias. - Futurity
How do we vet AI health tech? - Politico
AI's next frontier: Selling your intentions before you know them - TechXplore
How AI is polluting our culture - WBUR
Will AI revolutionize drug development? Maybe. - The Conversation
Somebody tell Hinton that midwit Harvard MBAs control genius MIT nerds every day. - the Guardian
Does investment research make sense in the age of AI? - FT
AI is a great sell-side analyst, herdy imitative skin-deep marketing fluff 🤣 . Those guys work hard and explain companies but don't really get paid to think outside the box or talk negatively. Great quarter guys!
AI technology credited for early detection of Orange County fire. - Yahoo News
Maybe they can invent an AI smoke detector that can tell if I'm just making pizza.
Film Forum screens movies that warned us about tech and AI through Jan 23 - Film Forum
25 experts predict how AI will change business and life in 2025 - Fast Company
A jailbreak that asks LLMs to rate the safety of their responses, and produce less safe responses? - The Hacker News
Follow the latest AI headlines via SkynetAndChill.com on Bluesky