PRESS PLAY
Weekly summaries of the AI news that matters!
Based on 6 episodes we listened to · September 2026
Last Week in AI connects headline developments to the infrastructure, incentives and policy tensions moving the AI industry.
Last Week in AI turns a dense news cycle into an extended conversation about the forces behind each announcement. Product launches share space with compute buildouts, chip supply, government policy and the commercial ties between model developers and cloud providers.
A Cybercab promise meets your next question: can Tesla actually remove the supervisor?
Press play if you want
Short reviews from the PlayNext desk, based on the episodes we processed.
A focused look at how export controls, chipmaking limits, and Huawei's designs are reshaping China's AI chip market.
China's AI chip market gives this episode a clear frame for looking at what export controls mean once the policy headline has passed. The conversation links Nvidia's shrinking lead with Huawei's gains, SMIC's capacity and yield constraints, and chip design choices that help…
A sharp look at AI safeguards, government contracts, and the uneasy logic of AI-driven self-improvement.
The episode tracks a live pressure point in AI policy: a proposed government contracting clause, national-security arguments involving China, and the unsettled question of what safeguards should apply to systems that may behave unpredictably.
A grounded look at how AI infrastructure deals can strain the alliances behind the scaling race.
The episode finds its tension in the uneasy bargain between AI model developers and the hyperscalers funding and supplying them. Its clearest thread is Amazon's deal with Anthropic: Amazon wants Anthropic pulled toward its own hardware and CUDA alternative, while the partnership…
A grounded look at AI platform access, safeguards, and the choices that shape real-world capability.
Last Week in AI frames a new chat interface as more than a product update, using it to examine how quickly safeguards can be tested once access expands.
A grounded look at how GPU clusters, wiring choices, and speed shape the frontier AI race.
The episode reads the frontier AI race through the hardware that has to exist before any grand claim can matter. Its center is xAI's fast buildout of an H100 training cluster, with the practical pressures around fundraising, power, wiring, and deployment speed treated as part of…
Tesla's Cybercab pitch meets valuation pressure, autonomy doubts, and the wider AI market story.
Tesla's Cybercab talk lands where product ambition meets the hard math of autonomy. The episode keeps its focus on what is already priced into the company's future, then grounds that question in concrete points: cybercabs discussed at $30,000 or less, a rough comparison with a…
A weekly panel of veteran gaming journalists breaking down game reveals, industry news, and layoffs.
Global tech with real-world consequences.
Hosts Andrey Karenkov and Jeremy introduce a weekly AI news episode, discussing personal updates and teasing major topics like ChatGPT Pro, Amazon Nova, GenFM, and Llama 3.3.
Andrei Kurenkov hosts with guest Gavin Purcell, who shares his background in media and AI startups, while they humorously mix up names and discuss upcoming AI news topics.
Hosts Andrey Kurenkov and Jeremy Harris introduce Last Week in AI, covering major news like Sora, Gemini 2, and OpenAI's AGI rule, while sharing personal anecdotes about moving and using AI audio tools.
["Anthropic's Project Glasswing and Claude Mythos dominate Last Week in AI #240, where the hosts say the model's offensive-cyber results are a major step up from Opus 4.6.", "Andrey Karenkov says he studied AI in grad school and works at Astrocade, while Jeremy Harris says he works in AI national security at CloudStone AI; they open by noting they skipped a week because of Andrey's birthday travel.", "They cite a browser exploit benchmark where Opus 4.6 found something bad in 14% of trials and Mythos reached 72%, plus a virology trial where the best Mythos run hit two critical mistakes.", "The rest of the episode moves through Google's Gemini 3.1 Flash Live, Anthropic's Cloud Code access changes, OpenAI's canceled adult mode, Microsoft's MAI models, Suno v5.5 customization, and Anthropic's $30 billion revenue run rate with a new Google TPU deal."]
["Last Week in AI #239 centers on OpenAI's Sora shutdown, Anthropic's desktop-control rollout, Cursor's Composer 2 controversy, and a Trump administration contracting clause aimed at AI vendors.", "The hosts are Andrey Karenkov, who studied AI in grad school and now works at Astrocade, and Jeremy Harris, who does AI national-security work at Gladstone AI.", "OpenAI is discontinuing Sora and its video API while Anthropic adds browser, mouse, keyboard, and display control to Claude, and Google tests Gemini task automation on Pixel 10 Pro and Galaxy S26 Ultra.", "Cursor's Composer 2, Adobe Firefly's custom-model beta, Luma AI's Uni1, and the GSA's proposed 'any lawful government purpose' language make the episode feel like an AI systems-and-policy roundtable rather than a pure model-news recap."]
Last Week in AI #190 centers on Bloomberg's report that OpenAI, Google, and Anthropic are all struggling to push their next models forward.
Last Week in AI #189 centers on OpenAI's predicted outputs feature, Anthropic's Haiku 3.5 price hike, and OpenAI's accidental o1 preview leak.
Last Week in AI #188 centers on OpenAI Search in ChatGPT, Apple Intelligence beta features, and GitHub Copilot adding Claude 3.5 Sonnet, Gemini 1.5 Pro, and OpenAI o1 models.
Last Week in AI #186 covers Adobe Max announcements including Firefly video generation, Project Scenic, Project Motion, Project Clean Machine, and Project Supersonic.
Weekly roundup and discussion of the most interesting AI news stories.
A weekly roundtable discussing the most interesting AI news from the past week.
A weekly roundup and discussion of the most interesting AI news, including research, apps, and listener comments.
A weekly roundup and discussion of the most interesting AI news, hosted by machine learning researchers.