🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

“Thinking is difficult, that’s why most people judge.” ― Carl Jung

There's a huge difference between knowing the name of something and knowing something. We talk in fact-deficient, obfuscating generalities to cover up our lack of understanding.

„In the age of AI, critical thinking must become the heart of the curriculum, and the heart of all society. It’s the filter that allows us to navigate a world where information is abundant…but meaning is contested. And it’s not a skill we’re born with“ psychologytoday.com/us/blog/all-th…

Happy 70th Birthday Tim Berners-Lee, the inventor of the World Wide Web! #InternetHistory

Cope. The fact that LLMs and LRMs can’t reliably reason has nothing to do with what is (or isn’t) scary. It has to do with how they are implemented. People who can’t be bothered to dig into how current models actually work—and what their actual limitations are—aren’t HELPING

AI can reason, in particular neurosymbolic AI. This paper (and many others) prove it: arxiv.org/abs/2212.12050 Reasoning in LLMs need to be engineered, not "evolved".

Current LLMs can't reason reliably... AI can!

There is a history of AI reasoning research, in particular #neurosymbolic AI in which for many years researchers have investigated sound, well-founded reasoning and learning. Time for reflection on the best of both approaches. Pascal Hitzler Moshe Vardi Francesca Rossi

If we are to accept Ilya Sutskever's premise that "digital computer" = "biological brain" then how are we going to scale up the GPUs to handle 100s of trillions of neurons, reduce the energy consumption to 0.3 kWh and fit the data centres in the skull?

Current business model pretends that software is free when in fact users pay with data. A new business model is urgently needed. Copyright law is not the problem.

Although the required number of moves grows exponentially with the number of discs in Tower of Hanoi, one would expect LLMs to handle 10 discs (~1000 moves, 50k tokens) but accuracy drops dramatically to near zero with only 10 discs.

There is no easy way to use current AI (LLMs) for marking (without generating very bad or silly mistakes).

OMG. The Apple paper “was written by an intern?” That’s what you have got? Anybody who says this is clueless. The paper was actually written by SIX people — including Yoshua Bengio’s brother. In science, the first author is *often* a junior author. In fields like biology and

An AI that does protein folding like a human would not be very useful or succesful, right? Time to talk about domain-specific AI vs. generalist AI.

finally a court gets serious about hallucinations:

Philosophers and economists, please, like Erik Brynjolfsson