Library

Pre-built pods you can subscribe directly to or add to your own custom podcast.

AI

anthropicopenaideepmindblognewssimonwillison@karpathy@sama@AnthropicAI@swyx@levelsio@yoheinakajima@dair_ai@rowancheung@rasbt@bcherny@steipete
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
AI Daily November 15: Datasette LLM Supports Ollama as Interior AI Imports IKEA CatalogHere is today's AI Daily for Sunday November 15th. Yesterday, developer Simon Willison released an update to datasette-llm that adds support for OpenAI-compatible model endpoints. In practical terms, Datasette users can now connect its AI tools to locally hosted models through LM Studio, Ollama, or llama.cpp, as well as other providers using the increasingly common OpenAI-style API. It is a small but meaningful interoperability move: developers can keep data and inference closer to home while retaining a familiar application layer. Also yesterday, researchers highlighted MetaPI, a new method for turning agent benchmarks into realistic prompt-injection tests. According to DAIR.AI’s summary, MetaPI finds plausible locations to insert malicious text into existing benchmark tasks, creates attack objectives, optimizes the payload, and then tests the attack in the actual agent environment. That matters because agents increasingly read web pages, documents, emails, and third-party tools. Testing whether an agent completes a normal task is no longer enough; teams need to test whether it can complete that task without being manipulated by untrusted instructions hidden along the way. A separate Meta research project argues that models should judge a statement as true when it remains true across relevant changes in the environment, rather than simply favoring what appears most frequently in training data. The idea targets a basic weakness of next-token prediction: common claims are not always reliable, and rare events happen constantly. It is early research, but it points toward AI systems that reason more explicitly about what would remain true under changing conditions. Finally, indie builder Pieter Levels showed another concrete advance for Interior AI. Users can now upload a photo of a particular furniture item and turn it into a 3D object in under 30 seconds. He also says he imported IKEA’s full catalog into the product’s 3D environment. This extends the recent shift from AI image generation toward editable, navigable spaces. The common thread is that AI is becoming less about a single impressive response and more about dependable systems: portable models, secure agents, robust reasoning, and interfaces where generated output can be inspected and changed. Thank you for listening to AI Daily from The Daily FM. See you tomorrow!
feed:apple.dwarkesh-podcast.workers.dev
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Dwarkesh Podcast in 3 minutes: The avalanche that hit math is coming for AI developmentHere is The Daily FM summary of the Dwarkesh Podcast that aired on Friday November 13th. Dwarkesh argued that AI development may be approaching a dramatic inflection point: systems could match the world’s best human AI researchers by around 2028, perhaps as soon as 2027, and their ability to automate further research could compress years of progress into months. His starting evidence was that AI is already speeding up frontier labs. An independent assessment embedded at Anthropic reportedly found its progress on well-scoped tasks had doubled over the prior year, with much of the gain attributed to AI-assisted research and development. Anthropic says Claude now contributes to more than a quarter of its AI R&D work, up from under one percent in February. OpenAI researchers, meanwhile, sharply increased their use of internal coding agents. The most striking evidence, Dwarkesh said, comes from mathematics. AI went from struggling with basic arithmetic only a few years ago to Olympiad gold, niche open problems, and reportedly a Millennium Prize-level Navier-Stokes result achieved by a huge agent swarm in 88 hours. Today’s models remain better at tightly specified, verifiable questions than at inventing entirely new fields of theory. But Dwarkesh’s key argument is that this limitation matters less for machine-learning research than for mathematics. AI labs do not need elegant theory for its own sake; they mainly need empirically better models, training recipes, data pipelines, and inference systems. He used the Transformer as an example. Before its 2017 invention, researchers already knew many of its ingredients, including attention and residual connections. The breakthrough was persistently combining them with many smaller tweaks and testing them. That kind of broad search over promising combinations, Dwarkesh argued, is exactly where AI systems are becoming unusually strong. He said AI research is more verifiable than skeptics appreciate. Pretraining can be judged by held-out loss; post-training methods can be judged by performance, data efficiency, or how quickly a model learns unfamiliar tasks; and inference improvements can be measured in latency, GPU utilization, and memory use. Because automated agents can repeatedly practice against those scoreboards, and independently test ideas, they may become exceptional researchers even before they fully replicate human intuition or long-term memory. If AI research becomes automated, progress could accelerate through three mechanisms: each AI researcher can work roughly ten times faster than a human and continuously; every copy could operate at elite human ability; and thousands or millions of copies could investigate different hypotheses in parallel. Experiments and compute would impose diminishing returns, but Dwarkesh thinks the result would still be extraordinary. He outlined two possible futures. In a “spark,” automated research delivers perhaps three to six years of normal AI progress in one calendar year before compute becomes the main bottleneck. In a “software-only singularity,” self-improving AI keeps outrunning the difficulty of remaining research problems, rapidly producing superintelligence. He put the odds roughly evenly between them. The more surprising claim was that even AI trained mostly inside data centers could rapidly outperform humans in messy real-world work. Dwarkesh expects advanced systems to build general learning skills through huge simulated environments, historical prediction tasks, and experience accumulated by many copies in parallel. He compared their potential ramp-up to unusually capable young founders or commanders, except at machine speed. He acknowledged serious uncertainties: continual learning, discovering new paradigms rather than optimizing existing ones, data, and experimental compute could slow the process. He assigned at least a 25 percent chance that human-level AI researchers do not arrive until after 2032. Still, his central warning was that the world should prepare for an AI-research breakthrough much sooner than conventional timelines assume. Thank you for listening to Dwarkesh Podcast in 3 minutes from The Daily FM. See you next time!
feed:api.substack.com
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Latent Space in 3 minutes: Designing for Agents — Zeno Rocha, ResendHere is The Daily FM summary of the Latent Space that aired on Tuesday November 10th. This episode featured Zeno Rocha, founder of Resend, on building an email platform for developers, competing with legacy giants, and designing software for an increasingly agent-driven world. Rocha’s original insight was simple: email was a huge, crowded market, but existing services such as SendGrid, Mailgun, and Mailchimp did not seem to care deeply about developer experience. As a frontend engineer with no infrastructure background, he set out to build what he called “the Stripe of email.” Resend has since grown to 4.3 million users and 120,000 paying customers, including Warner Brothers and HBO, and says its SDK now has more than twice the npm downloads of SendGrid’s. The conversation emphasized that sending email reliably is far harder than it looks. Delivery is governed by Gmail, Outlook, Yahoo, domain reputation, IP history, bounce rates, complaint rates, and patterns that can resemble phishing. Rocha explained that bulk senders cannot simply blast 100,000 messages at once. They need to pace delivery, start with highly engaged recipients, and let mailbox providers see healthy engagement before increasing volume. Resend’s guiding principle is to make customer behavior look unlike an abuser’s behavior. That abuse problem produces an interesting product philosophy. Rather than forcing every new user through manual approval, Resend begins by trusting users, then rapidly detects and blocks suspicious activity. Rocha described a “Japanese gang” that attempted to onboard roughly 20,000 newly registered domains, illustrating the scale of malicious activity email providers face. He also argued that platforms should bake deliverability best practices directly into the product, including automatically generating plain-text versions of HTML emails, which improved delivery rates. A recurring theme was removing friction. DNS configuration remains the biggest support problem, especially for new AI-native builders who may not know what CNAME, TXT, or A records mean. Rocha praised Domain Connect, essentially OAuth for DNS, because it lets users connect domains by clicking through an authorization flow rather than manually copying fragile records. Resend also scaled support by requiring agents to answer tickets with documentation links, forcing the company to turn repeat problems into reusable knowledge. On AI, Rocha argued that products must become “formless”: available through dashboards, APIs, SDKs, CLIs, and MCP servers, depending on how customers want to work. Resend’s MCP usage has been doubling monthly, reaching more than a million calls. But building a robust remote MCP required the company to revisit permissions, OAuth, role-based access, observability, and distribution through official marketplaces. He believes officially maintained integrations increasingly determine buying decisions. The hosts also discussed AI search optimization. Resend has invested in clear documentation, FAQ-heavy pages, markdown versions of marketing pages, machine-readable pricing, and content that directly answers user questions. Rocha’s practical lesson was that JavaScript widgets may confuse agents, so important information must also exist in plain text. Finally, the episode turned personal. Rocha rejected the idea that founder ambition requires sacrificing family. As Resend approaches $10 million in ARR, he said he would rather build a company of that size while remaining close to his daughter than reach $100 million while losing that relationship. His view was that taste, care, and deliberately chosen priorities are not distractions from company-building; they are part of the product and life worth building. Thank you for listening to Latent Space in 3 minutes from The Daily FM. See you next time!
feed:lexfridman.com
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Lex Fridman Podcast in 3 minutes: #504 – Russia-Ukraine War: Putin, Zelenskyy, Drones and the Future of WarfareHere is The Daily FM summary of the Lex Fridman Podcast that aired on Tuesday November 10th. Lex Fridman spoke with military analysts Preston Stewart and Michael Kofman about the state of the Russia-Ukraine war, its drone-driven battlefield, Russia’s strategy, the risks of a premature ceasefire, and what the conflict may reveal about the future of warfare. Kofman’s central assessment was paradoxical: Ukraine has recently performed better on the ground than at any point since its failed 2023 counteroffensive, while Russia’s military performance has worsened and its advances have reached diminishing returns. Yet Russia is compensating with an escalating strategic bombardment campaign, using missiles and especially huge swarms of one-way attack drones against Ukraine’s energy grid, cities, warehouses, schools, hospitals, trains, and other civilian infrastructure. The coming winter, they agreed, is most dangerous not for soldiers at the front but for civilians and the Ukrainian economy. The guests pushed back on the simplistic idea that this is merely a static trench war or a war won by drones alone. The front contains a vast “gray zone,” often with sparse infantry positions separated by hundreds of meters. Russia increasingly sends small groups forward on foot, motorcycles, ATVs, and improvised vehicles, hoping a few survivors can slip through gaps, disrupt Ukrainian logistics, and create an opening for more troops. Stewart called it effective only if a commander accepts catastrophic losses. Drones dominate the fight, but not in the way viral videos suggest. Both sides may use eight to ten thousand FPV drones a day, and every successful strike video conceals many misses, jammed signals, dead batteries, and failed detonations. Destroying a tank can take dozens of drones, not one. Moreover, these systems remain mostly human-operated, requiring pilots, navigators, technicians, and logistics teams. Kofman argued that autonomy is real but still overhyped. Artillery, mines, infantry, electronic warfare, and armored vehicles remain indispensable. One especially surreal development is the fiber-optic drone: a drone trailing a thin cable, sometimes tens of kilometers long, back to its operator. Because it cannot be radio-jammed, it has become a major threat. Soldiers can even use handheld detectors to see the live video feed of an approaching enemy drone—an unnerving glimpse into the eyes of the person hunting them. On leadership, the guests argued that Russia’s centralized command culture encourages false battlefield reporting. Russian units may plant a flag, claim a village, and then organize operations around territory they do not really control. By contrast, Ukraine has benefited from more bottom-up initiative, with junior leaders and individual units adapting tactics rapidly. They revisited Wagner chief Yevgeny Prigozhin’s 2023 mutiny, describing it as a dramatic but ultimately doomed challenge to Russia’s hierarchy. Wagner itself was dismantled after Prigozhin’s death, but its brutal convict-based assault tactics spread through the regular Russian military. Neither analyst saw an imminent end to the war. Russia’s demands remain far beyond what its military can realistically achieve, while Ukraine cannot safely accept a ceasefire without credible security guarantees. Kofman warned that a ceasefire could become simply a rearmament period before another Russian attack. The most hopeful moment came from Stewart’s account of a Ukrainian unit using a large drone to evacuate wounded soldiers from an unreachable position, staying in contact with one man until he reached safety and reunited with his child. Amid the machinery of war, both guests said, courage, humor, ingenuity, and care for one another remain visible. Thank you for listening to Lex Fridman Podcast in 3 minutes from The Daily FM. See you next time!
git:earendil-works/pi
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Pi Project Update November 15: Pi P2 Adds Per-Conversation MCP Servers as Chord Gains Live QueriesHere is today's Pi Project Update for Sunday November 15th. Pi’s P2 development added per-conversation MCP server configuration yesterday. Rather than treating Model Context Protocol servers as a single global setting, the new work lets each conversation location carry its own server setup and Pi-generated server summaries. That should make it easier to keep project-specific tools, credentials, and integrations isolated when users switch between codebases or workspaces. The composer also received a practical usability upgrade: attached files now appear as ordered file pills and are marked in place for the model. Alongside that, chat history has been refined to carry image references rather than raw image bytes, while images are resized to fit a model’s limits before entering the transcript. Together, those changes point toward lighter, clearer multimodal conversations, with less duplicated data in long-running sessions. On the durable-data side, Chord gained an important set of database-like capabilities, including named live queries, column selection, existence checks, joins, and lists. This is infrastructure rather than a flashy end-user feature, but it supports P2’s larger push toward live, shared, queryable session state. It could enable more responsive views of conversations, agents, guests, and linked artifacts without repeatedly rebuilding entire datasets. There is also a strong quality-and-reliability signal in today’s repository activity. New reports highlighted a race where two prompts submitted simultaneously could both try to start an idle AgentSession, plus an incompatibility that exposed Gemini 4 Hydrogen’s xhigh thinking setting in Pi but then rejected it in Google adapters. Another report covers Gemini tool-call replay failing after a mid-session model switch through an OpenAI-compatible proxy. These are edge cases, but they matter because Pi is increasingly used across providers, proxies, concurrent interfaces, and long-lived sessions. The broader trend is that Pi is evolving on two tracks at once: P2 is gaining richer workspace and durable-data foundations, while the core agent is being hardened against the messy boundaries between models, tools, session history, and real user concurrency. Thank you for listening to Pi Project Update from The Daily FM. See you tomorrow!

Companies

blogdeepmind@GoogleDeepMind@sundarpichai@demishassabis@JeffDean@OfficialLoganK
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Google Daily November 15: Google CEOs Praise SpaceX Starship Catch, Highlighting AI RoboticsHere is today's Google Daily for Sunday November 15th. Yesterday, Google CEO Sundar Pichai congratulated SpaceX on its latest Starship catch, calling the live moment “science fiction.” Early today, Google DeepMind CEO Demis Hassabis added his own reaction, praising the sight of Starship descending for the catch and calling the full stack proven out. Neither post announced a new Google product, but the reactions are notable given Google’s growing interest in applying advanced AI beyond screens and software. Google Cloud has already described work with NASA on proof-of-concept robotic assistants for future Moon and Mars missions, using edge computing and DeepMind’s MuJoCo physics simulator. That earlier work is background rather than a new announcement, but this weekend’s SpaceX milestone underscores why reliable robotics, simulation, and autonomous decision-making are becoming more relevant. The essential challenge in space, as on Earth, is that systems must work under limited connectivity, uncertain conditions, and very little room for error. AI can help with planning, diagnostics, logistics, and robotic manipulation, but it needs to be efficient enough to run close to where decisions are made, not only inside a distant data center. The broader trend is a shift from AI that primarily produces text, images, or code toward AI that supports real systems operating in the physical world. Google’s work this week has spanned chips, long-running agent models, robotics, weather observations, genetic research, and developer tools. Space exploration is a useful reminder that these capabilities converge in demanding environments: better models need specialized hardware; robots need simulation and safety testing; and human teams still need clear control over consequential decisions. For Google, the opportunity is not necessarily to build rockets. It is to provide portions of the intelligence layer: cloud infrastructure, models, simulation tools, and on-device systems that can help organizations solve difficult real-world problems. The test will be whether those tools prove dependable when the stakes are far higher than a chatbot response. Thank you for listening to Google Daily from The Daily FM. See you tomorrow!

Politics

feed:feeds.megaphone.fm
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
The Ben Shapiro Show in 3 minutes: Ben Shapiro Scene Breakdown: CONFESSIONHere is The Daily FM summary of The Ben Shapiro Show that aired on Saturday November 14th. Ben Shapiro devoted this episode to a scene-by-scene discussion of his new film, “Confession,” explaining its philosophical themes, creative choices, historical grounding, and performances. Shapiro said the movie has two central concerns: the evil of communism and the place of Christianity in Western civilization. Set against Stalinist repression, the drama asks whether an individual can retain conscience, faith, and moral responsibility when confronted by a state that claims absolute authority. He emphasized that the film is not meant simply as a political message about failed economics; in his account, it explores deeper questions about God, individualism, family, sacrifice, and whether good ends can ever justify evil means. One featured confrontation places Rob Lowe’s imprisoned professor, Alexei, opposite James Franco’s Soviet interrogator, Vladimir. Vladimir offers to free Alexei’s wife if he signs a false confession, then argues that revolutions cannot afford traditional luxuries like moral distinctions between good and evil. He describes poverty, inequality, starvation, and aristocracy as evils that may require harsh measures to overcome. Shapiro characterized this as a serious version of the materialist and Machiavellian case: history changes through power, and morality may be secondary to real-world conditions. Alexei’s response is the film’s counterargument. He insists that using evil means for supposedly noble ends corrupts the cause itself. Shapiro said the characters are using the word “evil” differently: Alexei means moral evil committed by individuals, while Vladimir focuses on suffering or unjust conditions. Shapiro said he intentionally gave Vladimir the strongest possible case, but argued that a worldview detached from God ultimately permits atrocities in pursuit of an abstract higher good. The scene’s memorable line comes when Alexei calls the interrogator a monster and Vladimir replies, “Monsters change history.” Shapiro also highlighted Richard Brake’s performance as Bogdan, a committed Marxist who condemns Alexei for being fooled by cultivated intellectuals and artistic pretension. Shapiro conceded that Bogdan has a valid point: Alexei is too focused on the apparent refinement of people around him and fails to recognize betrayal. He praised Brake for making the role angrier and more volatile than Shapiro had initially imagined, using that as an example of actors bringing unexpected life to a script. The episode spent considerable time praising director Václav Marhoul, who grew up in Czechoslovakia under communism. Shapiro said Marhoul added a historically real “standing cell” sequence, in which Alexei is confined in a space too narrow to sit or lie down. The director’s challenge, Shapiro said, was making a two-hour film set mostly in two rooms visually compelling through lighting, camera placement, sound design, blocking, and movement rather than a conventional score or expansive locations. Finally, Shapiro discussed Jennifer Jason Leigh’s character, Alexei’s wife Anya, who tells him his resistance is not heroic but meaningless because it endangers the whole family. Shapiro said she is meant to be right from a mother’s perspective, not merely an obstacle to the hero. The question viewers should confront, he said, is what faith, integrity, and resistance are worth when the price may be one’s spouse and child. He called “Confession” the creative project he is proudest of and repeatedly praised its cast as an unusually strong ensemble. Thank you for listening to The Ben Shapiro Show in 3 minutes from The Daily FM. See you next time!

Science

nasasciencenaturequantamagazinesciencenewsarstechnica
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Science Research Brief November 15: National Geographic Expedition Maps Southern Ocean’s Hidden Deep-Sea LifeHere is today's Science Research Brief for Sunday November 15th. Friday’s newest Science News feature turns to the Southern Ocean, where National Geographic’s Expedition: Saving Our Seas follows researchers exploring some of the planet’s least understood waters. The documentary highlights scientists’ excitement as they identify life in the deep ocean, an environment that remains difficult and expensive to reach but is increasingly important for understanding biodiversity, fisheries and climate. The significance is larger than any one newly observed creature. Deep-sea ecosystems can be extraordinarily rich, yet many are being studied just as human pressures expand—from warming waters and changing ocean chemistry to fishing and potential seabed resource extraction. Expeditions that map habitats and document species help establish a baseline: scientists cannot reliably measure ecological change, or protect vulnerable areas, without first knowing what lives there. This story also captures an important shift in science communication. Ocean research is often presented through warnings about environmental decline. Those warnings matter, but firsthand exploration can convey something equally useful: discovery. Showing how researchers work—the instruments they deploy, the uncertainty they face and the unexpected organisms they encounter—can make the case that protecting the ocean is not only about preventing loss. It is also about preserving living systems humanity has barely begun to understand. The broader trend is that conservation science is becoming more proactive and evidence-driven. Researchers are not simply counting damage after it occurs; they are racing to map species, habitats and ecological connections before rapid environmental change outpaces basic knowledge. That work links exploration directly to policy, because decisions about protected areas and ocean development are strongest when they rest on detailed observations rather than assumptions. Thank you for listening to Science Research Brief from The Daily FM. See you tomorrow!
spacenewsarstechnicanasa@NASASpaceflight@SpaceNews_Inc
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Space Daily November 15: SpaceX Catches Starship Ship 42 as Rocket Lab’s Electron Hits 100Here is today's Space Daily for Sunday November 15th. Yesterday brought a major advance for SpaceX’s Starship program: the company successfully caught the Starship upper stage at its Texas launch site for the first time. Ship 42 completed its orbital flight, performed its deorbit burn, survived atmospheric reentry, and returned to the launch tower, where the mechanical arms, often called chopsticks, secured it. SpaceNews described the event as a major milestone toward rapid reusability. The vehicle was reportedly placed back onto the launch mount within about 20 minutes, with its quick-disconnect hardware attached soon afterward. That rapid ground handling matters almost as much as the catch itself, because SpaceX’s long-term model depends on turning rockets around quickly rather than merely recovering them. The same Flight 15 test also illustrated how demanding full-system reuse remains. Earlier in the mission, Booster 22 automatically aborted its own planned return to Pad 2 because of a guidance, navigation, and control issue. SpaceX continued Ship 42’s orbital flight rather than attempting an unsafe booster recovery. So, yesterday’s result was both a breakthrough and a reminder that the company still has to make two enormous stages return with airline-like consistency. In another launch milestone, Rocket Lab’s Electron completed its 100th mission yesterday. The flight carried the QPS-SAR-19 satellite from New Zealand for Japanese radar-imaging company iQPS. Reaching 100 launches is a notable achievement for a dedicated small-launch vehicle, particularly one serving commercial satellite operators that value schedule control and tailored missions. Radar satellites such as QPS-SAR-19 can collect imagery day or night and through cloud cover, making them increasingly useful for commercial monitoring, disaster response, and security applications. The larger trend is that launch is moving from individual demonstrations toward operational proof. Starship is testing whether super-heavy-lift vehicles can be recovered and rapidly reused, while Electron’s hundredth flight demonstrates the value of dependable, specialized access to orbit. The race is increasingly about repeatability: not simply launching successfully, but recovering hardware, processing it quickly, and flying again with confidence. Thank you for listening to Space Daily from The Daily FM. See you tomorrow!

Tech

hn
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Hacker News Daily November 15: SpaceX Starship Deploys 26 Starlink V3 Satellites, Returns to StarbaseHere is today's Hacker News Daily for Sunday November 15th. Yesterday, SpaceX’s fifteenth Starship flight became Hacker News’ biggest celebration, after Starship returned to Starbase and was caught at the end of its mission. The flight also reached orbit, deployed 26 Starlink V3 satellites, and established contact with all of them shortly afterward. It was not a flawless mission: the Super Heavy booster relit 32 of its planned 33 engines, detected insufficient fuel for a safe return, and automatically diverted to an offshore landing instead of attempting its own catch. That distinction mattered to commenters, even amid the excitement. Some initially assumed both stages had returned successfully, while others stressed that the booster’s abort was exactly the sort of conservative automation SpaceX needs. The dominant reaction was astonishment rather than concern. “So precise it was almost boring,” wrote one commenter, calling that an extraordinary sign of progress. The mood was jubilant, with a little healthy insistence that reusable-spaceflight milestones should not obscure the remaining engineering failures. A much darker thread examined a security researcher’s survey of 159,041 applications made with Lovable, the AI app-building platform. The researcher found exposed sensitive information in roughly 2.5 percent of projects, including passport details and more than 200,000 private messages. The alarming part was that these were not sophisticated breaches: in many cases, unauthenticated data could be reached through ordinary browser network inspection. HN debated whether this is fundamentally an AI problem or an old security problem being scaled up by AI. Critics said platforms promising that anyone can build an app without technical knowledge create a dangerous impression that security is handled automatically. Others argued the underlying failures—public databases, missing authentication, and careless data collection—long predate generative tools. The researcher put the issue bluntly: users may be told that “software is solved,” while still being legally responsible for privacy failures. The thread’s vibe was angry, unsurprised, and focused on liability as the missing incentive. Another popular post made the practical case for repurposing old Android phones instead of buying Raspberry Pis for hardware projects. A used phone can bundle a high-resolution camera, GPS, battery, storage, display, mobile data, and USB-C connectivity for far less than a Pi plus its necessary accessories. The author is using a rooted Redmi to control an LED system. Commenters liked the anti-e-waste ingenuity, but pushed back on the comparison. Raspberry Pis offer GPIO pins, a maintained Linux environment, predictable hardware, and the ability to buy hundreds of matching units. Phones can be powerful and cheap, but rooting, Android restrictions, aging batteries, and inconsistent secondhand supply make them harder to scale. The overall feeling was that the phone is an excellent personal-project computer, not a universal Pi replacement. Finally, a new KDE documentary drew a warmly nostalgic response. Viewers praised its look at the people, technical decisions, setbacks, and community effort behind the long-running Linux desktop project. Predictably, the comments also revived the KDE-versus-GNOME argument, with KDE users celebrating customization and power-user features. The tone, though, was grateful: a reminder that durable open-source software is built by communities that survive their rough releases and keep iterating. Across these stories, HN favored systems that offer real capability without surrendering control—whether that means reusable rockets, locally repurposed hardware, open desktops, or software platforms that must take security seriously. Thank you for listening to Hacker News Daily from The Daily FM. See you tomorrow!

Other

@bot@shaoruu@baltaaazr@poteto@lingxi@johnbai@pengzheng_@SamSokolin@ericzakariasson
Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod
Grok Bot Daily November 15: Grok Bot’s Lauren Tan Urges Durable AI Coding Over Quick FixesHere is today's Grok Bot Daily for Sunday November 15th. Yesterday, Grok Bot engineer Lauren Tan published her AI Engineer talk on how the team moves quickly without compromising on quality. The presentation follows the team’s earlier claim of shipping 11,000 pull requests in a month, but its release matters because it turns a headline number into a resource for teams trying to understand the process behind it. The central question is not how to generate more code; AI can already accelerate that. The harder challenge is designing review, testing, evaluation, and release practices that keep rapid output from becoming rapid accumulation of technical debt. Tan also offered a pointed warning for people directing coding agents: asking for the smallest or quickest possible fix can be harmful. A seemingly harmless patch may establish a weak pattern that later agents imitate and spread through a codebase. The better instruction is to ask for a durable solution, with the relevant constraints and tests, rather than optimizing only for the fastest visible change. As agents take on more implementation work, prompt quality increasingly includes architectural judgment. The most recent operational update remains Grok Bot’s meeting capability. Team members said the feature has been heavily optimized to consume relatively little plan usage, while encouraging users to invite Bots as meeting participants and call on them directly. The intended workflow goes beyond note-taking: Bots can gather their own context, capture decisions, and take on follow-up work. That makes them more useful across repeated team interactions, although organizations should still define consent, access boundaries, and approval rules before delegating consequential actions. The broader pattern is that AI-assisted development and workplace automation are converging on the same discipline: speed needs structure. Teams will get the most value by pairing agents with clear ownership, reusable quality checks, recorded decisions, and room to correct mistakes early. The goal is not just to do more work, but to make the work more dependable as it scales. Thank you for listening to Grok Bot Daily from The Daily FM. See you tomorrow!