Hey! I'd love to hear your thoughts, send me a voice note.

Artificial intelligence can explain quantum physics, write software, analyze thousands of documents, and reason through complex problems.

Then you start a new conversation...

and it may not remember what happened yesterday.

So what does it actually mean for an AI to remember?

In Episode 10 of Zero to Singularity, we explore one of the most important problems standing between today’s AI assistants and truly persistent intelligent agents: long-term memory.

We break down the difference between a model’s learned parameters and its context window, why a huge context window is not the same as permanent memory, and how modern AI systems use retrieval, embeddings, vector databases, summaries, and external memory stores to preserve information across time.

Then we go deeper.

How should an AI decide what is worth remembering? How does it know when a memory is outdated? What happens when two memories contradict each other? Can an AI learn from experience without retraining its entire neural network? And how can it keep learning without suffering catastrophic forgetting gaining new knowledge while damaging what it already knows?

We explore:

context windows, working memory, episodic memory, semantic memory, retrieval-augmented generation, vector databases, agent memory, continual learning, catastrophic forgetting, memory compression, personalization, privacy, memory poisoning, and the possibility of AI systems that accumulate experience over months or even years.

Hey! I'd love to hear your thoughts, send me a voice note.

Modern AI was built on an extraordinary resource: human knowledge.

Books. Websites. Research papers. Code. Images. Video. Conversations. Billions upon billions of examples created by people.

But what happens when frontier models have already consumed most of the useful, accessible human-generated data?

In Episode 9 of Zero to Singularity, we investigate one of the biggest questions facing the next generation of artificial intelligence: Can AI keep improving if high-quality human training data becomes a bottleneck?

This episode explores:

• how pretraining data powers modern AI
• scaling laws and diminishing returns
• the difference between more data and better data
• synthetic training data
• self-play and automated curriculum generation
• reinforcement learning and verifiable tasks
• model-generated reasoning examples
• simulation and virtual environments
• multimodal data from video, audio, robotics, and sensors
• proprietary and expert-generated datasets
• inference-time scaling and test-time compute
• retrieval, tools, agents, and external memory
• model collapse and synthetic-data contamination
• why AI-generated internet content could become a training problem
• whether machines can eventually generate their own useful learning experiences

We also examine some of the biggest claims surrounding the future of AI training:

Is the internet running out of useful data? Can synthetic data replace human-created knowledge? Does training on AI-generated content inevitably cause model collapse? Can self-play generate effectively unlimited training material? And could future AI systems eventually design their own increasingly difficult curriculum?

The deeper question is no longer simply:

How much human knowledge can AI absorb?

It may become:

Can intelligence eventually create the experiences it needs to make itself smarter?

Hey! I'd love to hear your thoughts, send me a voice note.

What happens when artificial intelligence stops living only on screens and starts acting in the physical world?

In Episode 8 of Zero to Singularity, we explore Physical AI the convergence of advanced AI, robotics, and embodied intelligence.

We break down how modern robots are learning to see, understand instructions, manipulate objects, move through unfamiliar environments, and connect language with physical action. We also examine why teaching a machine to operate safely and reliably in the real world may be far harder than teaching an AI to generate text, code, images, or plans.

This episode explores:

• embodied AI and robotics foundation models
• vision-language-action systems
• humanoid robots versus specialized machines
• reinforcement learning and imitation learning
• teleoperation, simulation, and synthetic data
• dexterity, locomotion, spatial reasoning, and force control
• the robotics data problem
• why impressive demos do not automatically equal reliable deployment
• safety, recovery, and human supervision
• the economics of physical automation
• factories, warehouses, and the much harder challenge of the home
• what Physical AI could realistically look like by 2030

The central question:

Was language actually the easy part?

Because in software, intelligence only has to produce the right answer.

In robotics, intelligence has to survive reality.

Zero to Singularity explores artificial intelligence from the fundamentals to the frontier making complex technology understandable without oversimplifying the science.

Hey! I'd love to hear your thoughts, send me a voice note.

What happens when AI can do the work that used to train beginners?

In Episode 7 of Zero to Singularity, we explore one of the most important questions in the future of work: whether artificial intelligence is beginning to reshape the traditional career ladder from the bottom up.

The episode examines why entry-level workers may be feeling the effects of AI before the broader workforce, how reduced hiring can matter just as much as layoffs, and what happens when routine junior work is automated before young professionals have a chance to build experience.

We break down:

• the difference between a task, skill, job, and occupation
• automation versus augmentation
• why junior workers may be affected differently from senior workers
• the growing “apprenticeship problem”
• codified knowledge versus tacit knowledge
• how AI can both replace beginner tasks and accelerate learning
• changes in software, finance, law, customer support, and creative work
• the rise of AI agents and longer autonomous workflows
• why verification, judgment, and domain expertise may become more valuable
• whether AI could broaden jobs instead of simply eliminating them
• how education and early-career training may need to change
• six possible futures for work through 2030

The central question is simple:

If AI removes the first rung of the career ladder, how do humans learn to climb?

Zero to Singularity explores artificial intelligence from the fundamentals to the frontier making complex ideas understandable without oversimplifying the science, economics, or uncertainty.

Hey! I'd love to hear your thoughts, send me a voice note.

What happens when artificial intelligence begins helping build the next generation of artificial intelligence?

In Episode 6 of Zero to Singularity, we explore one of the most consequential ideas in modern AI: automated AI research and recursive self-improvement.

We break down the difference between an AI correcting one answer and an AI actually becoming more capable. From self-refinement and synthetic data to automated coding, AI scientists, verifiers, model training, and research agents, this episode follows the increasingly automated pipeline behind AI development.

We explore questions including:

• Can AI identify its own weaknesses?

• Can AI design and test improvements to other AI systems?

• What is the difference between self-correction and true self-improvement?

• Why are verifiers so important?

• Can AI-generated training data eventually degrade future models?

• What happens when an AI learns to improve the process that improves AI?

• Could AI research eventually accelerate faster than human-led research?

• And what would actually have to happen before we could call it recursive self-improvement?

The episode also examines the limits: compute, energy, hardware, scientific judgment, reward hacking, model collapse, diminishing returns, and the continued role of human researchers.

The central question:

What happens if the best AI researcher in the world eventually becomes an AI?

Zero to Singularity explores artificial intelligence from the fundamentals to the frontier making complex ideas understandable without oversimplifying the science.

Hey! I'd love to hear your thoughts, send me a voice note.

Does AI actually understand the world or is it just getting incredibly good at predicting patterns?

In Episode 5 of Zero to Singularity, we explore one of the biggest questions in modern artificial intelligence: whether AI systems are beginning to build internal world models representations of space, objects, cause and effect, and what might happen next.

We break down, in plain language:

• what a world model actually is

• the difference between predicting words, video frames, and the consequences of actions

• what “latent space” means and why it matters

• how JEPA-style systems try to model reality without generating every pixel

• whether language models develop internal maps of space and time

• what Othello-GPT revealed about hidden internal representations

• why AI still struggles with physics, object permanence, and long-horizon prediction

• what “simulation drift” means

• whether AI needs a physical body to truly understand cause and effect

• how world models could shape robotics, self-driving cars, autonomous agents, and AGI

The deeper question is this:

If an AI can predict what will happen in unfamiliar situations, plan around those predictions, and act successfully in the world, when do we stop calling it pattern matching—and start calling it understanding?

Zero to Singularity explores artificial intelligence from the fundamentals to the frontier, making complex AI concepts understandable without oversimplifying the science.

Research current through September 2026.

Hey! I'd love to hear your thoughts, send me a voice note.

What would actually have to happen before we could honestly say AGI has arrived?

In Episode 4 of Zero to Singularity, we explore the scientific, technical, and economic debate around Artificial General Intelligence and why there is still no universally accepted test for it.

We break down the strongest proposed signs of general intelligence: reasoning, memory, continual learning, autonomy, generalization, metacognition, tool use, world models, physical understanding, and the ability to adapt to unfamiliar problems. We also examine why high benchmark scores do not automatically prove general intelligence, especially when tests can become saturated, contaminated, scaffolded, or gamed.

The episode also tackles the biggest unresolved questions:

• Does AGI need a body?

• Does it need consciousness?

• Could an AI become economically “general” before becoming cognitively human-like?

• Are today’s models truly learning general skills—or becoming extraordinarily good at familiar kinds of tests?

• Could AGI arrive gradually enough that nobody agrees on the exact moment it happened?

We compare competing paths toward AGI, from scaling current models and inference-time reasoning to world models, embodiment, neuro-symbolic systems, and self-improving AI.

Zero to Singularity explores artificial intelligence from first principles to the technological frontier separating evidence from hype and asking what the next stage of machine intelligence would actually look like.

Hey! I'd love to hear your thoughts, send me a voice note.

What happens when an AI model is given more time and compute to reason before answering?

In Episode 3 of Zero to Singularity, we explore modern reasoning models, including inference time compute, chain-of-thought, reinforcement learning, self-correction, backtracking, hidden deliberation, and why more “thinking” does not always produce better answers.

We also examine the core debate: Are these systems genuinely reasoning, or are they producing increasingly sophisticated statistical simulations of reasoning?

Featuring research and examples from OpenAI, Anthropic, Google DeepMind, DeepSeek, Qwen, and the broader AI research community.

Research current through September 2026. This episode contains AI-generated audio.