Machine Learning Street Talk (MLST) reposted this
A great discussion by two very competent scientists (Tim Scarfe and the one and only Dr. Keith Duggar). Brilliant discussion. I'm with "the (biomachine) - the Duggar" all the way 🙂
MLST is a media production company based in London focussing on producing science-based content. Our associated AI podcast is the top-rated globally on Spotify. Each week we have hard-hitting discussion with the leading thinkers in the AI space. Please get in touch if you are interested in sponsoring the show in some way, or need AI advisory work.
External link for Machine Learning Street Talk (MLST)
London, GB
Machine Learning Street Talk (MLST) reposted this
A great discussion by two very competent scientists (Tim Scarfe and the one and only Dr. Keith Duggar). Brilliant discussion. I'm with "the (biomachine) - the Duggar" all the way 🙂
Machine Learning Street Talk (MLST) reposted this
The Machine Learning Street Talk (MLST) team had an interesting discussion about whether ChatGPT o1 is actually reasoning. I often find myself disagreeing with Dr. Keith Duggar about individual points that he makes but I mostly tend to agree with his conclusions. There's something like reasoning going on here but it's a pretty shallow reasoning. I had similar thoughts about how well these models seem to "understand": https://lnkd.in/gZkpZahc It's an interesting watch: https://lnkd.in/gD-isSpv
Saurabh Baji discusses Cohere's approach to developing and deploying large language models (LLMs) for enterprise use. We just dropped the interview on MLST. Some key points: Cohere creates AI models for practical business use rather than trying to make the largest ones possible. Their AI can work in the cloud or on a company's own computers and can use a company's data safely. These models can be adjusted to work better for specific tasks. Cohere has grown a lot, now working with many large enterprise customers. Oracle and TD Bank use Cohere's AI for various tasks including HR and finance. Cohere's SVP expects many more companies to start using AI seriously in the next year or two. He believes it's crucial that this AI is trustworthy, safe, and reliable.
Machine Learning Street Talk (MLST) reposted this
AI's Future: 2040 Pedro Domingos & Tim Scarfe on AI Technological Innovations and Future Directions, its regulation, and its potential to amplify collective intelligence. Pedro challenges the idea that AI should be heavily regulated, comparing it to regulating mathematics or quantum mechanics. According to Domingos, the real danger isn't AI becoming too smart but rather it being too *stupid*—making poor decisions due to a lack of common sense. "Anybody who calls an LLM a stochastic parrot is just revealing their ignorance of machine learning 101" (1:56:58) https://lnkd.in/eEvKGUVx Machine Learning Street Talk (MLST) Moving beyond the current hack-based approaches in AI development, he emphasizes the need for a unified master algorithm that combines diverse paradigms—such as deep learning, symbolic AI, and geometric deep learning—into a more robust foundation. He introduces the concept of Tensor Logic, which aims to bridge the gap between neural networks and symbolic AI, potentially revolutionizing how we approach AI tasks. Moving on, he then discusses Symmetry-Based Learning, which could unlock new, undiscovered principles in AI, leading to systems that are more efficient and capable. He also dives into how AI can enhance societal decision-making, moving beyond outdated political systems to create real-time, collective intelligence. Imagine a society where the intelligence of the whole is exponentially greater than the sum of its parts! - AI regulation could stifle innovation rather than protect us. - The real power of AI lies in amplifying both individual and collective intelligence. - We need a unified approach in AI development, merging diverse paradigms for a stronger foundation. - The future of AI lies in discovering new symmetries, not just utilizing existing ones. - Tensor Logic could be the key to unifying neural and symbolic AI. - We need to shift from hacks to systematic, formal foundations in AI development. #AI #MachineLearning #Innovation #CollectiveIntelligence #Technology #FutureOfAI #PedroDomingos #TensorLogic #GeometricDeepLearning
Machine Learning Street Talk (MLST) reposted this
AIs are not Turing Machines, beautifully and simply explained by Dr. Keith Duggar . Turing Machines are simply Finite State Machines with an unbounded memory. LLMs are also Finite State Machines but current training methods make it impossible for the AI to break out of the Finite State Automata boundary and into the space of Recursive Grammars.
Machine Learning Street Talk (MLST) reposted this
Principal Data Scientist | Co-author of Machine learning for High-Risk Applications | Kaggle Grandmaster(Notebooks)
Machine Learning Street Talk (MLST) has been dropping some seriously cool interviews back-to-back which I’ll highly recommend: 1️⃣ Prof. Subbarao Kambhampati spilling the tea on ChatGPT’s reasoning skills (or lack thereof). 2️⃣ Sayash Kapoor on AI existential risk and government policy 3️⃣ Sara Hooker discussing two crucial topics: • a) Why measuring AI risk by raw compute power (FLOPs) is overly simplistic. Spoiler: It's far more nuanced than that. •b) The AI Language Gap aka why our AI assistants are English snobs and the implications for global AI accessibility. Props to MLST and Tim Scarfe for these insightful interviews. 📺 :https://lnkd.in/gtVJdscr #MachineLearning #AI #
Machine Learning Street Talk (MLST) reposted this
Had a fun interview on Machine Learning Street Talk with Tim Scarfe on "Is ChatGPT an N-gram model on steroids?" Title based on my recent work in understanding transformers in terms of n-gram statistics and Subbarao Kambhampati's (another recent MLST guest) claim https://lnkd.in/e7SBKTki Episode link: https://lnkd.in/ew8kyJgd Paper + discussion: https://lnkd.in/eXJgNcsz
Machine Learning Street Talk (MLST) reposted this
Scientist, author (5 books, including Rebooting AI (Forbes 7 Must Read Books About AI), and Founder (Geometric Intelligence. Acquired by Uber, and Robust.AI). Professor Emeritus, NYU.
What has and has not changed in AI since the ChatGPT revolution? Full video of a talk I gave Friday, with thanks to Machine Learning Street Talk (MLST). Also included in link below is a commentary by Ben Goertzel, AGI pioneer.
We filmed Gary Marcus keynote at AGI-24. We also recorded a 2 hour interview with him which we will release soon. Here are the best bits.
We spoke with Jay Alammar from Cohere about retrieval augmented generation and how Cohere is using LLMs to solve real-world business problems. On MLST now! You should all check out Jay's new book coming out too "Hands-On Large Language Models: Language Understanding and Generation" --> https://amzn.to/3WGv4XZ