The Neuron: AI Explained

Diffusion models changed how we generate images and video—now they’re coming for text.

In this episode, we sit down with Stefano Ermon, Stanford computer science professor and founder of Inception Labs, to unpack how diffusion works for language, why it can generate in parallel (instead of token-by-token), and what that means for latency, cost, and real-time AI products.

We talk through:
The simplest mental model for diffusion: generate a full draft, then refine it by “fixing mistakes”

Why today’s autoregressive LLM inference is often memory-bound—and why diffusion can shift it toward a more GPU-friendly compute profile

Where Mercury wins today (IDEs, voice/real-time agents, customer support, EdTech—anywhere humans can’t wait)

What changes (and what doesn’t) for long context and architecture choices

The real-world way to evaluate models in production: offline evals + the gold-standard A/B test

Stefano also shares what’s next on Mercury’s roadmap—especially around stronger planning and reasoning for agentic use cases.

Try Mercury + learn more: inceptionlabs.ai

For more practical, grounded conversations on AI systems that actually work, subscribe to The Neuron newsletter at https://theneuron.ai.

Customer service is one of the industries most impacted by AI — but what if AI alone isn’t the answer?

In this episode of The Neuron Podcast, Grant Harvey and Corey Noles sit down with Matt Price, Founder & CEO of Crescendo, to explore how AI and humans working together can outperform automation alone. After spending 13+ years at Zendesk, Matt is now building an AI-native customer experience platform that automates up to 90% of tickets with 99.8% accuracy — without sacrificing empathy, trust, or outcomes.

We cover:
• Why LLMs are the biggest shift in customer service since the telephone
• Why bolting AI onto old CX workflows fails
• How Crescendo’s multimodal AI can chat, talk, see images, and control devices in one conversation
• Real-world examples (like smart sprinkler troubleshooting via voice + vision + APIs)
• Why Crescendo combines AI agents with forward-deployed human experts
• How outcome-based pricing aligns incentives around real customer satisfaction
• How AI is reshaping (not eliminating) customer service jobs
• Why “deflection” is the wrong mindset for CX — and what replaces it
• What customer support roles look like in an AI-native future

This is a deep dive into the next generation of customer experience, where AI handles scale and speed — and humans deliver judgment, empathy, and innovation.

Subscribe for weekly conversations with the builders shaping the future of AI and work.

Subscribe to The Neuron newsletter for more interviews with the leaders shaping the future of work and AI: https://theneuron.ai

Taylor Mullen, Principal Engineer at Google and creator of Gemini CLI, reveals how his team ships 100-150 features and bug fixes every week—using Gemini CLI to build itself.

In this first in-depth interview about Gemini CLI's origin story, we explore why command-line AI agents are having a "terminal renaissance," how Taylor manages swarms of parallel AI agents, and the techniques (like the viral "Ralph Wiggum" method) that separate 10x engineers from 100x engineers. Whether you're a developer or AI-curious, you'll learn practical strategies for using AI coding tools more effectively.

🔗 Links:
• Gemini CLI: https://geminicli.com
• GitHub: https://github.com/google-gemini/gemini-cli
• Subscribe to The Neuron newsletter: https://theneuron.ai

In this week's live-stream replay, we go live for a 2-hour, hands-on deep dive into GPT-5.1 Codex Max with Alexander Embiricos, product lead for OpenAI Codex. You’ll walk out feeling like an agentic-coding wizard, even if you’re starting from zero. GPT-5.1 Codex Max is OpenAI’s latest frontier agentic coding model. It’s built on an upgraded reasoning backbone and trained to handle real-world software engineering tasks end to end: PRs, refactors, frontend builds, and deep debugging. It can work independently for hours, compacting its own history so it can refactor entire projects and run multi-hour agent loops without losing context. In this live session, we’ll set it up together, build real agents, and push Codex Max to its limits.

Modern AI has been dominated by one idea: predict the next token. But what if intelligence doesn’t have to work that way?

In this episode of The Neuron, we’re joined by Eve Bodnia, Founder and CEO of Logical Intelligence, to explore energy-based models (EBMs)—a radically different approach to AI reasoning that doesn’t rely on language, tokens, or next-word prediction.

With a background in theoretical physics and quantum information, Eve explains how EBMs operate over an energy landscape, allowing models to reason about many possible solutions at once rather than guessing sequentially. We discuss why this matters for tasks like spatial reasoning, planning, robotics, and safety-critical systems—and where large language models begin to show their limits.

You’ll learn:
What energy-based models are (in plain English)

Why token-free architectures change how AI reasons

How EBMs reduce hallucinations through constraints and verification

Why EBMs and LLMs may work best together, not in competition

What this approach reveals about the future of AI systems

To learn more about Eve’s work, visit https://logicalintelligence.com.

For more practical, grounded conversations on AI systems that actually work, subscribe to The Neuron newsletter at https://theneuron.ai.