🔥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

[2304.09276] Towards a Neural Lambda Calculus: Neurosymbolic AI Applied to the Foundations of Functional Programming arxiv.org/abs/2304.09276

We are deeply saddened by the passing of Bill Atkinson. He was a true visionary whose creativity, heart, and groundbreaking work on the Mac will forever inspire us. Our thoughts are with his loved ones.

Nicola Saccomanno Mehrdad Farajtabar Neural nets can be engineered to perform sound reasoning in various logics up to a level of expressiveness at which more research is needed: arxiv.org/abs/2012.05876

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

On the current AI debate on reasoning; neurosymbolic AI offers useful insights and sound techniques.

In this paper we show that “reasoning is not all you need in AI, but it could get some attention”. Neurosymbolic AI: the 3rd wave | Artificial Intelligence Review ⁦Artur d'Avila Garcez⁩ link.springer.com/article/10.100…

Neurosymbolic AI is the answer to LLMs inability to reason: theconversation.com/neurosymbolic-…

LLMs' "illusion of thinking" paper highlights a limitation of chain-of-thought prompting that I elaborated on in: staff.city.ac.uk/~aag/papers/AI…

Reasoning in neural nets needs to be engineered, not "evolved". Correct reasoning is achievable by using architecture and loss function, not prompting and hoping for the best.

Integration of learning and reasoning in neural nets should produce model compression rather than requiring more and more data and scale.

Great piece by Artur d'Avila Garcez on neurosymbolic AI. Worth reading to reflect on the current state of AI research.

Long before LaTeX, most mathematical theses relied on hand-inserted symbols. But in 1968, Dennis Ritchie - creator of C - typeset his entire thesis, subscripts and all, with astonishing accuracy. How exactly he pulled it off remains a mystery. Learn more in this week's paper

François Chollet Free energy is used in this paper to add a reasoning layer to deep learning: arxiv.org/abs/2505.20313

Gary Marcus Demis Hassabis I had the privilege of attending a NeurIPS Conference workshop (many years ago now) where Demis Hassabis presented AlphaZero results and debated with Gary Marcus on issues of transfer and out-of-distribution learning that are still relevant today. We need more debate at that level!

A Neural Lambda Calculus Bridging the gap between neurosymbolic AI and the foundations of computing through Neural Lambda Calculus. goatstack.ai/topics/a-neura… #NeurosymbolicAI #LambdaCalculus #DeepNeuralNetworks #ComputationalModels #goatai #goatstack via GoatStack.AI

Terence Tao. Worth listening to.

Carl is right. AI is already boosting productivity in many tasks, but the real power of general-purpose technologies, from the steam engine and electricity to AI, lies in catalyzing new products and processes.

I just signed up to volunteer for 2025 Seattle Marathon Summer 5k/10k/Half Marathon and Kids Fun Run using Race Roster. Join me raceroster.com/events/2025/90… via Race Roster