Podcast
The Neuron: AI Explained
Hosted by The Neuron · EN
The Neuron is a daily newsletter with 700,000+ readers that covers the latest AI developments, trends and research; this is our podcast, hosted by Grant Harvey and Corey Noles. We aim to create digestible, informative and authoritative takes on AI that get you up to speed and help you become an authority in your own circles. Available Wednesdays and Sundays on all podcasting platforms and YouTube.
Subscribe to our newsletter: https://www.theneurondaily.com/subscribe
42episodes
Episodes
Newest firstAll episodes
BONUS: Building Real-Time AI Voice Agents with LiveKit's Ben Cherry
2w ago01:09:02Tap to summarizeVoice agents are moving from “cool demo” to real product infrastructure.In this livestream, we’re joined by Ben Cherry of LiveKit to break down what it actually takes to build real-time AI agents that can listen, respond, interrupt, call tools, and work in production.LiveKit is an open source framework and developer platform for building voice, video, and physical AI agents in production.We’ll talk through the stack behind real-time AI experiences, then build and test a live demo together on The Neuron.In this live demo, we’ll cover:🎙️ How LiveKit helps developers build voice, video, and physical AI agents⚡ What makes real-time agents different from normal chatbots🧠 How voice agents handle latency, interruptions, speech, and tool calls🛠️ Why production-ready AI agents are much harder than a weekend demo🚀 What builders should know before shipping voice AI to real usersAnd yes, we’re doing a live demo, which means there is at least a small chance the agent talks back at exactly the wrong time. Perfect television.Guest: Ben Cherry, LiveKitLiveKit: https://livekit.com/Ben on LinkedIn: https://www.linkedin.com/in/bcherry-product-engineerBen on GitHub: https://github.com/bcherrySubscribe to The Neuron for clear, useful AI news, demos, and explainers for people trying to understand where this tech is actually going.https://www.theneuron.ai/
Transcribe →The AI Trying to Solve Math’s Biggest Mystery w/ Tudor Achim of Harmonic
2w ago00:46:34Tap to summarizeWhat happens when AI stops simply giving answers and starts producing proofs a computer can verify?In this episode of The Neuron, Corey Noles and Grant Harvey talk with Tudor Achim, Co-Founder and CEO of Harmonic, the company behind Aristotle — a formal reasoning system built to generate machine-checkable mathematical proofs. Tudor explains why math may be the clearest test case for moving AI from “trust me” to “check me,” and why formal verification could matter far beyond Olympiad benchmarks.They discuss what “mathematical superintelligence” actually means, why Tudor thinks solving a Millennium Prize problem would be a meaningful threshold, and how Lean-based proofs could change the way mathematicians collaborate. They also explore Aristotle’s real-world use cases, from open math problems to verified software, chip design, scientific computing, and the future of AI-assisted discovery.Plus: why Tudor thinks formal math has reached a “zero to one” moment, why specs may be the bottleneck in verified software, and why humans still need to direct the questions AI systems try to solve.Subscribe to The Neuron and sign up for The Neuron Daily at theneuron.ai.
Transcribe →BONUS: Can AI Actually Be Your Therapist? We Ask the CEO Building One
3w ago01:13:55Tap to summarizeMay is Mental Health Awareness Month, and as AI becomes more embedded in our daily lives, one of the biggest questions we face is whether these systems can responsibly support emotional and psychological well-being.AI chatbots are increasingly being used for emotional support, but recent lawsuits faced by OpenAI and earlier ones targeting character.ai and Google's AI Overviews, as well as clinical reports, and internal research have raised valid concerns about their impact on vulnerable users.What does it take to build an AI system specifically designed for mental health from the ground up? Is that even possible?In this LIVE episode of The Neuron Podcast, Corey Noles and Grant Harvey speak with Daniel Reid Cahn, co-founder and CEO of Slingshot AI, about Ash, an AI application purpose-built for therapeutic support. Slingshot has raised $93M from a16z, Radical Ventures, and others to develop a foundation model for psychology trained on structured therapeutic conversations across modalities such as CBT, DBT, and psychodynamic therapy.We discuss the limitations of general-purpose chatbots in mental health contexts, recent controversies surrounding AI and psychiatric risk, and what differentiates a system designed to provide structured therapeutic engagement compared to one being used in a way it was never intended to be. The conversation also explores a broader question: Can AI meaningfully expand access to high-quality mental health care, and where should clear boundaries remain? Or should we keep our counseling where we always have, on a couch with a box of Kleenex and a hug nearby?🔗 Try Ash:https://www.talktoash.com/📌 About The Neuron PodcastThe Neuron breaks down the biggest stories in AI for 580,000+ daily readers. Our podcast goes deeper with the leaders, founders, and researchers shaping the future of AI. New episodes every week.Subscribe to The Neuron newsletter — theneuron.ai
Transcribe →Inside Genspark: $0 to $250M ARR in 12 Months with Wen Sang
3w ago00:44:38Tap to summarizeGenspark went from AI search startup to autonomous AI agent platform, hitting $250M ARR in 12 months with no paid ads until they bought a Super Bowl spot. Co-founder and COO Wen Sang joins Corey and Grant to explain what "AI employee" actually means, demos Genspark Claw live (including buying us coffee mid-interview), and lays out his big thesis: legacy software is becoming infrastructure while AI agents become the new interface between humans and work. We get hands-on with Workspace 4.0, Claw, and a custom agent built live for the show.• Genspark Workspace 4.0 announcement: https://www.genspark.ai/blog/genspark-ai-workspace-4• Genspark sb-git: https://genspark.ai/sb-git/intro• OpenAI's customer story on Genspark: https://openai.com/index/genspark/• Forbes AI 50 (2026): https://www.forbes.com/lists/ai50/• Marc Benioff on Salesforce Headless 360 (referenced by Wen): https://x.com/Benioff • Andrej Karpathy's "wiki for agents" idea (referenced as inspiration for sb-git): https://x.com/karpathy• Wen on the DealMaker Show: https://alejandrocremades.com/wen-sang/Try Genspark for free: https://genspark.aiSubscribe to The Neuron newsletter: https://theneuron.ai
Transcribe →BONUS: The AI Starter Kit: What to Try...and What to Ignore
4w ago02:03:15Tap to summarizeNew to AI and not sure where to start?Join us live Thursday for The AI Starter Kit: What to Try...and What to Ignore.This beginner-friendly session will help you cut through the noise and focus on the AI tools, habits, and prompts that actually matter. By the end, you’ll know what to try first, what not to worry about yet, and how to ask better questions when you get stuck.In this session, we’ll cover:🚀 The best first steps for AI beginners🛠️ What tools and features are worth trying now🙅 What you can safely ignore for the moment💡 Simple ways to get better answers from AI🔍 How to troubleshoot when AI gives you something unhelpfulWhether you’re brand new to AI or still figuring out how to use it well, this live session will give you a practical place to start.
Transcribe →Can AI Really Design New Drugs? Google DeepMind Spin-out Isomorphic Labs Explains
4w ago00:42:28Tap to summarizeCan AI move from predicting proteins to actually designing new drugs? Isomorphic Labs is trying to answer one of the biggest questions in science.In this episode of The Neuron, Corey Noles and Grant Harvey talk with Rebecca Paul, Head of Medicinal Drug Design at Isomorphic Labs, and Michael Schaarschmidt, Foundational AI Research Lead.They explain why drug discovery is so slow, expensive, and failure-prone—and why AI drug design is much more complicated than “generate a molecule and ship it.” The conversation covers AlphaFold, structure prediction, molecule generation, binding models, clinical failure rates, human trust in AI systems, and the long-term hope of designing drugs for targets once considered “undruggable.”In this episode:Why drug discovery can take more than a decadeWhat people misunderstand about “AI-designed drugs”How medicinal chemists actually use AI modelsWhy biology is harder than text, images, or codeWhat it would take to make drug discovery faster and cheaperThe dream of designing a drug candidate in one iterationWhy “undruggable” proteins may not stay undruggable foreverAdditional resources:Technical report blog Best resource for learning about the capabilities that we are buildingIsomorphic Labs websiteBest destination for learning more about Iso and joining our team in London, Lausanne or Cambridge, MASubscribe for more grounded conversations on how AI is changing science, work, and the world.For more practical, grounded conversations on AI systems that actually work, subscribe to The Neuron newsletter at https://theneuron.ai.
Transcribe →BONUS: OpenAI Workspace Agents 101: Build, Run, and Scale AI Workflows
May 101:24:03Tap to summarizeJoin us Thursday as we break down OpenAI’s new Workspace Agents and what they mean for the future of work.We’ll cover:⚙️ What workspace agents are🤖 How they differ from regular chatbots🏢 Where they fit into real team workflows🚀 How to start working with them effectively🔄 What agentic AI means for workplace automation📈 Why teams are shifting from one-off prompts to repeatable AI-powered processesWhether you’re experimenting with ChatGPT at work, leading AI adoption, or trying to understand where OpenAI is taking agents next, this session will help you see what’s possible and what to watch for.Tune in for a practical, hands-on deep dive into the future of AI at work.Sign up for The Neuron newsletter: https://www.theneuron.ai/
Transcribe →How Google's New AI Turns Anyone Into a Music Producer (Flow Music Demo)
Apr 2900:38:58Tap to summarizeGoogle just acquired an AI startup that lets anyone create real music, music videos, and custom instruments — no experience required. In this hands-on episode, Corey sits down with Kendall Rankin from Google to demo Flow Music (formerly Producer AI), the generative music tool now living inside Google Labs. They build a garage rock song about AI from scratch, generate a music video with VEO, and dig into what "amplifying human creativity" actually looks like when the tool can do most of the lifting. Listeners walk away with a clear view of where AI music tools fit in an artist's workflow, why watermarking (SynthID) matters, and how to try it for free.Try Flow Music: https://producer.ai Google Labs: https://labs.google SynthID (watermarking): https://deepmind.google/technologies/synthid/ Subscribe to The Neuron newsletter: https://theneuron.ai
Transcribe →BONUS: GPT 5.5 LIVE - The New GPT "Spud" Model is Here; Let's Break It
Apr 2501:39:35Tap to summarizeOpenAI dropped GPT-5.5, so we did the only reasonable thing: went live immediately and tried to break it.In this off-the-cuff Neuron Live, Corey and Grant walk through OpenAI's GPT-5.5 release notes, benchmark claims, rollout details, and early access reactions before testing the model live across coding, reasoning, creativity, web research, and absurd prompt challenges. We also compare a few GPT-5.5 responses against Claude Opus 4.7, test Codex, build a new version of Cat Doom, and ask the important questions, like whether a sentient vending machine that only dispenses expired tuna salad deserves to live.In this episode, we cover:• What OpenAI says is new in GPT-5.5• GPT-5.5’s improvements in coding, computer use, research, and knowledge work• Early benchmark results across Terminal-Bench, GDPval, Frontier Math, BrowseComp, and scientific research tasks• Why token efficiency may matter as much as raw intelligenceGPT-5.5’s rollout across ChatGPT, Codex, Plus, Pro, Business, and Enterprise• Live Codex testing with a one-shot Cat Doom game buildCreative stress tests involving palindromes, time-traveling potatoes, dystopian vending machines, and Lord of the Rings product reviews• First impressions of whether GPT-5.5 feels meaningfully different from GPT-5.4 and Claude Opus 4.7This was not a formal benchmark. It was a first-contact livestream: messy, fast, weird, and exactly the kind of test we like.Subscribe for more AI breakdowns, live model tests, beginner-friendly explainers, and weirdly useful prompt experiments from The Neuron.Sign up for The Neuron newsletter: https://www.theneuron.ai/Follow along for more AI news, analysis, and live experiments.
Transcribe →BONUS: LIVE: Claude Opus 4.7 Just Dropped. Here's What Actually Changed.
Apr 1701:01:31Tap to summarizeGrant and Kyle dive into a comprehensive review and live test of the newly released Claude Opus 4.7, a cutting-edge large language model. This session explores its capabilities for coding and game dev, specifically referencing the "Renaissance / Plan Final Fantasy Tactics RPG Game" project. Discover how this ai model performs under pressure and its potential impact on game design workflows.🔴 LIVE at 9:30AM PT / 12:30PM ETAnthropic just dropped Claude Opus 4.7, and we’re putting it through the gauntlet in real time.Join Grant Harvey (Lead Writer at The Neuron) for an unscripted, warts-and-all test of Anthropic’s newest flagship model.What we’re testing- Advanced coding on tasks Opus 4.6 struggled with- New higher-resolution vision support for images up to ~3.75 megapixels- File system-based memory across multi-session work- The new xhigh effort level, which sits between high and max- Claude Code’s new /ultrareview slash command- Auto mode for longer, less-interrupted agent runsWhy this mattersOpus 4.7 is the first model Anthropic is releasing with its new automatic cyber safeguards, following last week’s Project Glasswing announcement.It’s also the direct upgrade path from Opus 4.6 at the same price:- $5 per million input tokens- $25 per million output tokensIf you build on Claude, this is likely the model you’ll be using next.What’s changing under the hood- New tokenizer, where the same input can map to more tokens depending on content type, roughly 1.0x to 1.35x- State-of-the-art score on GDPval-AA, a third-party evaluation of economically valuable knowledge work- Better instruction following, which means prompts written for earlier models may now behave differently- Improvements across finance agent evals, document reasoning, and long-context tasksBring your hardest prompts. We’ll run them live and show you what breaks, what shines, and whether it’s worth migrating today.Watch part two, where Grant covers Codex for (almost) anything: https://youtube.com/live/OiRkwm3-og0📰 Full writeup in tomorrow’s newsletter: 🐱 Subscribe to The Neuron (700K+ readers): https://www.theneuron.ai
Transcribe →