Conversations that connect human judgment, creativity and context with AI’s capabilities. Find a voice, question an idea, and bring something useful back to your work.
These are AI Chemistry listening lenses, not fixed labels or endorsements. Every mode can learn from every show.
The complete collection
41 podcasts. Your next perspective.
Showing 41 of 41 podcastsYour shortlist stays in this browser when local storage is available.
Research & frontiersArchitect
Dwarkesh Podcast
Dwarkesh Patel
Long-form conversations about capabilities, research and the choices shaping its future. Patel gives guests room to unpack a forecast. A can show what a system does today, while the leap from that score to a prediction about tomorrow rests on assumptions about and .
Conversations connecting research with science, philosophy and the human experience. Fridman’s interviews put technical accounts of alongside arguments about consciousness. The distinction matters when a machine produces a convincing answer and people disagree about what it understands.
engineering discussions about models, and the infrastructure behind applications. The engineering trade-offs become tangible when an agent calls a tool. A fast demo can become a slow application once each request triggers several rounds of .
Founder and investor perspectives on companies, markets and where value may emerge. Guo and Gil look at where a startup can keep an advantage as improve. A feature that is easy to copy faces different economics from a product embedded in a customer's daily .
Deep conversations about capabilities, , safety and emerging applications. Labenz gives particular attention to the gap between an impressive capability and dependable behavior. Evaluations help make that gap visible by checking how a system performs when the task or instructions change.
Technology reporting on how enters products, public debate and everyday life. Roose and Newton follow the consequences of product decisions for the people using them. takes on a concrete meaning when a familiar app starts using personal conversations to tailor its responses.
Short-form analysis to make sense of developments and their implications for work. Whittemore puts announcements in context for people deciding what to try at work. A useful distinction is between gaining access to a new and changing a task enough for a team to adopt it.
Conversations with builders about products, research and company development. The founders’ perspective connects a research decision to a customer problem. matters here because the examples available to a team shape which requests its product can handle well.
Interviews and reporting on the companies and decisions shaping the technology industry. Kantrowitz examines the incentives behind product launches and platform decisions. That lens helps explain how distribution and pricing can influence as much as a technical breakthrough.
Technical perspectives on open models, post-training and learning from human feedback. Lambert digs into what happens after a base model is trained. Human feedback supplies examples of preferred answers, and reveals whether those preferences carry over to unfamiliar tasks.
Research and practitioner interviews connecting ideas with applications. Charrington invites researchers to explain how they tested an idea. The choice of often determines whether a reported gain is relevant to the application an engineer actually wants to build.
Extended discussions of research papers, reasoning, intelligence and the limits of . The hosts spend time on what an experiment actually demonstrates. A discussion of reasoning might turn on whether an rewards a memorized pattern or requires the system to solve an unfamiliar problem.
Conversations with researchers and founders about products and emerging businesses. Effron approaches research through the decisions a founder faces. A team choosing between an existing and a specialized one must weigh control over its product against the cost of maintaining it.
An investor and builder lens on infrastructure, markets and company strategy. The conversations connect infrastructure choices with business strategy. For a service billed per request, the cost of affects both the price a founder can offer and the customers it can serve.
Researcher and industry interviews that place developments in a broader context. Smith draws out the technical idea behind a research result. Following the path from a paper to a product makes it easier to see where new or a different test would be needed.
A research-informed roundup of developments across the landscape. The weekly format puts releases and papers side by side. Two announcements about can sound similar even when one changes who can use a system and the other changes what it can do.
Applied discussions about building, deploying and using in real systems. Benson and Whitenack bring deployment details into the conversation. becomes especially useful after launch, when the inputs arriving from users begin to differ from the examples used in development.
Conversations about possible futures, technology and opportunities for human progress. Hoffman and Finger invite guests to imagine how a technology could change a profession or a community. Their scenarios give listeners a way to consider which decisions still call for as capabilities expand.
Practitioner perspectives on the engineering decisions behind production . Biewald’s interviews offer a view from the team running the experiments. Tracking and comparing runs makes it possible to trace an apparent improvement to a change in the model, the examples or the test itself.
Business-oriented analysis of developments and their implications for teams. Roetzer and Kaput connect industry developments to decisions managers face. A change in an AI tool can affect the approval step in a marketing as well as the time it takes to produce a draft.
Hands-on demonstrations of how people use in practical . Vo walks through the steps behind a finished result. Seeing the original , the material a person supplied and the edits they made afterwards helps explain why the same tool can yield different outcomes.
An enterprise perspective on news, research and implementation decisions. The panel weighs technical announcements against enterprise constraints. A proposed looks different once a team accounts for access to internal records and who will handle exceptions.
A look at research through science, and frontier-model development. Fry asks researchers to unpack the experiment behind a result. Robotics makes the challenge especially clear because an action that works in simulation still has to succeed amid the variation of the physical world.
Organizational stories about leading, governing and scaling . The organizational cases follow adoption beyond the pilot. Decisions about who owns the , who can change a process and who signs off on an output help explain why the same tool spreads differently across teams.
In-depth discussions of , governance and decisions with long-term consequences. Guests work through the assumptions behind long-range decisions. In AI safety, an estimate of risk can change substantially depending on how soon a capability arrives and how long institutions have to respond.
Product and growth conversations, including how changes the work of product teams. Rachitsky’s product interviews connect customer discovery to what a team ships. For an AI feature, depends partly on whether users can recognize a poor answer and recover without losing their work.
Reporting and conversations on China, technology, chips and geopolitics. Schneider puts technical capacity in its industrial and political context. Access to chips constrains , which in turn shapes which organizations can afford large training runs.
Policy conversations on governance, national security and economic competition. The policy lens connects a technical capability to the institutions responsible for it. Rules governing access to , for example, can affect research collaboration as well as competition between countries.
Conversations about how builders, writers and operators use in their work. Shipper and Quintero explore the choices behind creative and operational work. The revealing part of a writing is often the moment the author rejects a fluent draft because it misses the intended argument.
Human-centered conversations about and its impact on business and daily life. El Kaliouby foregrounds the people who design and encounter these systems. A discussion of becomes more concrete when listeners can identify whose experiences are represented in the training examples.
Engineering conversations about software teams, development practices and coding tools. Orosz looks at how tools change the way engineers collaborate. When code generation speeds up a first draft, can become the point where a team spends most of its effort.
Accessible conversations and practical ideas for developing an everyday habit. Wilson starts from tasks that recur during the working day. Reusing a for meeting notes, for example, makes it easier to see which instructions help and which details still need checking.
news and commentary with a focus on media, technology and public life. Howell and Jarvis bring a media perspective to technology news. Questions of become tangible when a tool can reproduce the style of a creator whose work helped make the tool useful.
Practitioner conversations about operating and systems reliably. The engineering stories focus on what happens after a system goes live. gives a team a way to detect changing inputs and decide whether to investigate, roll back or retrain.
Conversations on , and the systems needed for dependable . Bowne-Anderson explores how engineers tell whether a system is working. With agents, a useful evaluation looks at the sequence of tool calls as well as the final answer, since an early mistake can propagate.
science conversations covering skills, tools and practical applications. Krohn’s interviews connect technical methods to the problems people use them to solve. A prediction becomes useful only when the data, the target and the decision someone will make fit together.
A critical-thinking lens on statistics, and the evidence behind claims. Polich examines how a claim survives contact with evidence. A result that looks impressive against one can weaken when the comparison changes or the sample leaves out an important group.
Conversations connecting with creativity, society and human choices. Sidhu explores the human choices behind a technical possibility. In creative work, generative AI raises practical questions of authorship when the concept, the draft and the final selection come from different contributors.
Narrative experiments exploring what happens when enter everyday life and work. Ratliff uses his own experience as the setting for the experiment. When an agent speaks or acts on someone’s behalf, the reactions of the people on the other end reveal how much trust depends on knowing who is present.
An accessible, entertaining take on news, tools and demonstrations. Pereira and Purcell give listeners a feel for tools through demos and commentary. Watching a turn into an image or a clip makes the gap between an entertaining result and a repeatable technique easier to spot.
Conversations about applications, and industry use cases. The industry examples connect computing hardware to the task it enables. A system that processes sensor in real time faces a different constraint from one that can spend hours training before it produces a result.