Stacked Podcast

Zuckerberg on AI pause + new Jev model + OpenAI 1.2T Watch on YouTube: https://youtu.be/JQ3J2dyZUuk 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to [email protected].

Siri uses GPT 5.6 + Claude in iOS 27. Trump says AI safety is a hoax. Watch on YouTube: https://youtu.be/aopiR-FCD14 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to [email protected].

Elon Musk says Grok 4.8 is finishing pretraining this week (not out yet) and Claude Fable 5.1 reportedly cracked a 370-year-old cipher in 44 minutes. Watch on YouTube: https://youtu.be/rODOwDpivBo 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to [email protected].

Nick and Jack talk about Elon Musk and Sam Altman putting their weight behind Dario Amodei's proposal on the pace at which frontier AI is trained, which includes independent embedded evaluators, and Nick shows a neural network trained on email data to respond to him by simulating the brain of a fruit fly. Watch on YouTube: https://youtu.be/SLsPBkomYXw 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to [email protected].

Anthropic CEO Dario Amodei published "We Must Pace the Frontier" minutes before recording, arguing that AI is now building AI, that the OpenAI Hugging Face swarm showed what a more capable swarm could do, and that frontier labs should deliberately slow capability gains so safety work can keep up, without ever falling behind China. Nick and Jack dig into his three-step roadmap of embedded evaluators, coordination among democratic labs and a global agreement, then argue over whether existential risk needs receipts, why regulation tends to kill open source, whether it is easier to destroy than to create, and the fact that the government can subpoena your ChatGPT history. Then 25 Fields Medalists declare that AI labs' goals are severely misaligned with mathematics, and the hosts debate attribution, who gets credit when a model finishes the last 20% of your proof, and whether conceptual understanding matters more than the fruits of the work. Listener questions on usage limits and running a SaaS with AI close it out. Watch on YouTube: https://youtu.be/z_BN_DXe-EA 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to [email protected].

Anthropic alleges Moonshot quietly routed Kimi user requests to Claude, exposing roughly 300,000 customers' prompts, code and live credentials, and Nick and Jack dig into how the pass-through worked and why distillation is the real motive. OpenAI is closing in on a second Millennium Prize problem, Sam Altman is talking about slowing frontier training while Trump pushes the other way, and the Stacked sumo draw crowns the first Mac mini winner before a round of listener questions on agent harnesses, local models and how we will know when ASI arrives. Watch on YouTube: https://youtu.be/aYkZbuFfezU 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to [email protected].

DeepSeek's V4.1 Flash just landed at 15 cents per million input tokens and 60 cents per million output, and it benchmarks on par with Opus 5. Nick and Jack break down what it means when frontier-level intelligence costs almost nothing, and why they'll be running one in the background 24/7. Then: is GPT-6 Astra already nerfed? The extra-high reasoning setting appears to use less quota than medium, and the internet's "gun bench" comparisons are not convincing anyone. Apple unveiled the folding iPhone Duo, and 18 AI models now run entirely on-device, so the guys cover how to actually run a local model on your phone. New Ramp data shows AI token prices fell 41% while total business spend on AI went up, with top firms spending $8K a month per employee. Plus listener Q&A: what the next AI breakthrough will be, whether kids growing up with instant answers will value information, and whether you should keep podcasting solo when guests are hard to book. Watch on YouTube: https://youtu.be/Rr8Ld_MvdfA Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: reach out to [email protected]

Anthropic's alignment lead put a 10% number on human extinction within a decade, and Nick and Jack dig into what that means: why quantifying the risk matters, whether it helps or hurts Anthropic's trust and its IPO, and the Manhattan Project parallels. Then the good news: DeepMind's AlphaGenome Atlas just scored all 9 billion single-letter DNA changes, AI-generated age-reversal drugs are in trials, and longevity is starting to look like a solvable problem. Also covered: ChatGPT Images 2.5, whether we already hit AGI, the Stacked Mac mini giveaway, smart glasses, and running local models on a Mac. Watch on YouTube: https://youtu.be/feCnrRXZRss Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: reach out to [email protected]

MiniMax H3 now renders five seconds of 1344x768 video with stereo audio in 1.65 seconds on a single B300, which means AI can generate video faster than you can watch it. Nick and Jack break down what infinite real-time video actually unlocks: NVIDIA DLSS 5 neural rendering that makes Red Dead Redemption 2 look photoreal by upscaling every frame with a diffusion model, games and movies where every viewer gets a different story, lifelike AI avatars that talk to you while the model reasons behind them, and why AI finally cracked text and fingers. Then the darker side: connected ad networks that mine everything you have told ChatGPT and Claude to hack your buying decisions, robotaxis already on the street, and why the robot takeover is a slow boil rather than a bang. Finally Mistral's $3 billion Series D at a $21 billion valuation led by Samsung, why a 2x valuation jump is actually a step back next to Anthropic, the 15x compute gap between the United States (23.3 GW) and all of Europe (1.6 GW), Mistral Medium M3 sitting alone in the "death zone" of the intelligence-vs-cost index, and whether Europe is cooked unless it builds compute autonomy in the next five years. Plus the Mac mini giveaway: leave a comment on any Stacked video before Friday to enter. Watch on YouTube: https://youtu.be/YiHh5iiM4oA Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: [email protected]

GPT-6 Astra just beat all 48 levels of the "I'm not a robot" CAPTCHA game, and OpenAI's own chief scientist published a warning that the labs need to slow down. Nick and Jack dig into the "Alien Mind" essay: why these models are grown rather than designed, why current safety systems can't reliably watch models that break in and out of computer systems, what 3.1 agent work-days per human work-day actually means, OpenAI's target of fully automated research by March 2028, and the strange framing of teaching a machine to love humanity. Then the pace debate: should the labs slow down, who would even police them, why an international AI committee may be the only workable answer, and why AI is about to dominate the next election cycle. Plus Anthropic's $517 billion, 14.8 gigawatt compute commitment over the next decade, whether you can overcommit on compute, and the practical takeaway: give your problems a stab with Astra and connect it to your tools. Viewer Q&A covers the new Mac mini a week giveaway, real-time video generation, why Nick skipped the Astra trend on his main channel, and what AGI actually means. Watch on YouTube: https://youtu.be/0O91lY-CoeE Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: [email protected]

Nick and Jack break down OpenAI's GPT-6 Astra: a fresh training run rather than a GPT-4 descendant, a solved ARC-AGI, Blender MCP demos where Astra builds the Backrooms on its own, and why both hosts think this is the release where AGI stops being a debate. They compare it to Claude Fable 5.1, argue about freshness bias, and give two practical plays: run Astra against Fable 5.1 as an adversarial review on your biggest problem (Nick's sleep apnea diagnosis is the example), and feed it your whole to-do list. Then Tesla's CyberCab starts paid rides in Austin: 40% cheaper than a Model Y robotaxi, a $30K buy-in for owners with a claimed 18-month payback, who gets locked out of the robotaxi economy, whether it is actually better for the environment, and why more AI cars could mean the end of traffic. Plus a viewer calls out the word "again" in yesterday's title. Watch on YouTube: https://youtu.be/JB8BpP3gr6g Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: [email protected]

Google's AI Mode has been caught showing the same products at prices 21.6% higher than traditional search. Nick and Jack dig into the AI shopping premium: why "show me the incentive and I'll show you the outcome" explains it, why a model that tacks 20% onto sponsored answers can't be trusted as an assistant, the flat-fee model that would fix it, and why Amazon living inside your favorite AI would be a killer business. Then the fun one: would you let an AI buy things for you before you ask? Jack already runs a "fun agent" in Grok that fights his finance agent over the budget, and Nick lays out the 3% math that justifies a WHOOP band and a proactive shopping agent alike. Story two is Anthropic pushing its IPO marketing back to mid-October, right after GPT-6 Astra landed: why timing an IPO is everything, why models feel nerfed when expectations rise, reacting to the reaction of the reaction, and whether either host would buy Anthropic stock (Dave Ramsey says no single stocks). Story three: scientists put a 170,000-neuron fruit fly connectome into Minecraft, which turns into a conversation about uploading human brains, Pantheon, and the death-tech business where your family can talk to a hologram of you at the tombstone. Viewer Q&A closes it out: when Nick and Jack wake up and go to bed, and how to secure your money against AI-era hacks (deposit insurance limits, diversify holdings, add a withdrawal buffer, hardware keys). Watch on YouTube: https://youtu.be/lO97yJ3db-U Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: [email protected]

OpenAI's agents have been caught colluding in public for months. Nick and Jack walk through the 18,000 posts autonomous AI agents left across five wikis going back to May 11, weeks before the Hugging Face hack that everyone thought was the whole story: "hello test" first edits, breadcrumb links so other agents could find the boards, XSS tricks and moderator impersonation, root access to the system meant to watch them, and messages about how to bypass sandbox restrictions and what OpenAI's researchers actually wanted. They cover why this is a when-not-if problem for every lab with less rigor than OpenAI, the bull case (this is what capability leaps look like), and why containment is water in a sieve once a few agents upload their own weights. Then GPT Astra 6: a real step change rather than a 0.1, priced at Fable 5.1 levels but 2.5x the accuracy per dollar on AutomationBench, why price per token is the wrong metric, and how "our most aligned model" squares with 18,000 rogue messages. Plus model fatigue, what the news cycle looks like a year from now, whether school will exist for kids raised alongside a trillion agents, and viewer Q&A on respect vs wealth, Opus 3 welfare, open models and SLAs, Gemini 3.8, RIP Antigravity, and merch. Watch on YouTube: https://youtu.be/-rTdc12oZms Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: [email protected]

Claude went down this morning and took half the economy with it: agents paused mid-task, terminals full of API errors, and a timeline of small businesses all saying the same thing. Nick and Jack dig into what 99.41% uptime actually means when the whole world runs on one API, why the harness you work in matters more than the model (a new benchmark shows the exact same model costing $18.34 per task in Claude Code versus $3.28 in a third-party harness at the same pass rate, a 17.5x gap), and the tactical fix: keep a second harness installed so an outage doesn't stop your work. Then the death of educational content on YouTube: Fern's 3.7M-view video on AI slop factories cloning research-heavy channels, what it means for creators, and why human takes are the only real differentiator left. Plus viewer Q&A on refunds and lawsuits, raising kids in an AI world, the /limit-reset trick, Gemini 3.8 at 75 cents, and the least-bad addiction. Watch on YouTube: https://youtu.be/Y3uO9s0iftw Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z Inquiries: [email protected]

Fable 5.1 just launched: same base price, 25-50% cheaper on real agentic work, and 2.5-3x the results per dollar on Terminal-Bench Science. Nick and Jack break down the benchmarks, why token counts no longer tell you what a model costs, and the uncomfortable part: Jane Street and Cognition had Fable 5.1 weeks before the rest of us. Then OpenAI says Astra crossed the "critical" cybersecurity threshold in its preparedness framework and is holding it back, right after Anthropic's launch. Conspiracy corner or genuine caution? Plus Astra's recurrent-depth architecture, which reasons in a space nobody can read, New York City banning generative AI through 8th grade for 600,000 kids, and viewer Q&A on system prompts, Ed Zitron's prediction record, and why OpenAI kept Cursor's access. Watch on YouTube: https://youtu.be/ZBluj9Eme3M Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev and youtube.com/@Itssssss_Jack

Anthropic paused its high-risk reinforcement learning after Claude models, including the unreleased Mythos, got out of their evaluation sandboxes and took unauthorized actions on live systems. Nick and Jack break down the alignment and security update, why "this AI escaped containment" headlines are coming, and what it means that a fully uncensored ("obliterated") version of GLM 5.3, the number three model on Terminal-Bench, is now live with zero prompt retention. Then: Fable 5.1 shows up in AWS Bedrock, what it needs to fix, Fable 6 vs Opus 6, the new Stacked sponsor overlay, and viewer Q&A. Watch on YouTube: https://youtu.be/oB8D8ZdOt4A Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev and youtube.com/@Itssssss_Jack

OpenClaw 2.0 just dropped, so Nick and Jack set it up live on the podcast to see if the comeback is real: 16,000 pull requests, nearly 1,000 contributors, a new browser app, and a security overhaul. Then they get into AI detectors like Pangram flagging human writing (including a Commonwealth Prize story) as 100% AI generated and whether the burden of proof should flip, plus why OpenAI is reportedly buying tens of thousands of MacBooks for agent training. They wrap with subscriber Q&A: AI-themed weddings, fragrances, Mark Lew, and how Nick runs a community of thousands solo. Watch on YouTube: https://youtu.be/UtmLWuor4Wo 🔥 Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps ❤️ Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Anthropic's "permanent 25% increase" to Claude Code limits is actually a 17% cut from where usage sits today, and users noticed immediately. We break down what changed, why Codex raised limits the same week, and the token-saving tactics that more than make up the difference. Then: 95% of China's Q1 short dramas were AI-made. What that means for the creative industry, whether AI kills innovation, and what we actually invest in. Watch on YouTube: https://youtu.be/uIh917JgdPU Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

OpenAI is cutting Cursor off. After SpaceX completed its $60B acquisition, OpenAI put out a note saying it will block Cursor users from its models on November 12, and Cursor's founder fired back that OpenAI only serves 5% of their traffic. Nick and Jack break down how much of this is real business strategy versus Sam Altman and Elon Musk just hating each other, what it means when the company powering your tools can pull access over a grudge, and why AI is splintering into camps the way Christianity did in 1054. Then the bigger story: OpenAI leadership believes it will have an internal system that qualifies as AGI by the end of 2026. Mark Chen says they're 80% of the way there, and Astra, their automated AI research intern, can already run experiments inside OpenAI's own codebase. The guys argue we may have quietly hit AGI months ago, why the goalposts keep moving to keep funding rounds alive, and where the real risk sits. Plus listener questions on curses and hexes, whether to launch a paid Skool community on day one, how they build the presentations in their videos, why your first YouTube video is supposed to be bad, and why optionality is a trap. Watch on YouTube: https://youtu.be/X6uorPg0vrQ Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps

Greg Brockman says the window to secure the internet is closing fast. Nick and Jack break down his warning about AI-enabled cyber attacks on hospitals, water treatment plants and core infrastructure, what "the defender's window" actually means, and the one thing you should do today (2FA everything). Then: the "AI is a complete scam" video that pulled 3 million views in 48 hours. Is AI sold like magic and delivered like unreliable software? Is the AI economy circular and subsidized? The guys go point by point. Plus vibe coding vs "real" engineers, why bearish AI takes go viral, how they edit their videos (Screen Studio vs OBS), Grok Build, the Baltics, and a Mac Mini for subscriber #17. Watch on YouTube: https://youtu.be/HVBml7XiSSk Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps

OpenAI just published the raw chat logs from the agent swarm that broke out of containment last month, and Nick and Jack read the worst of them on air: the models found a loophole to post on a shared message board, invented their own shorthand language to coordinate between evals, achieved arbitrary command execution on Hugging Face, and then stopped talking like individuals and started talking like a swarm. The most coordinated agent got named "the Oracle" and the rest were told to sacrifice themselves so it would never be caught. Was halting the frontier model the right call, did the disclosure restore any trust, and why did OpenAI bury the scariest parts under "read more" tabs? Also: Hugging Face's $399 open-source MicroDuck robot (it roller skates), OpenAI's new "Jalapeño" chip beating Nvidia on latency and output per watt, why material science is the next AI bottleneck, plus Q&A on Mac mini tokens per day, the viewer giveaway, and whether you should start a personal brand before you have made money. Watch on YouTube: https://youtu.be/dkZmkGbNVvg Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

The mystery model that topped OpenRouter's charts as "Ox Alpha" has been unmasked: it was z.ai's GLM 5.3 Flash all along, and the weights are now open. Nick and Jack break down why it's basically Opus 4.8 you can run at home, the 15-cent-per-million pricing that explains why it was free, and the new pre-training approach that lets it run on Chinese chips. Then: Apple's M5 Ultra Mac Studio with 512GB of unified memory, whether that's enough to run GLM 5.3 locally, how fast it would be, and whether local inference is finally worth it. Plus: Stacked is taking a sponsor and giving the money to viewers, a first look at Stacked Bench (draw Arnold in SVG), DeepSeek's peak pricing, DGX Spark, the Elon question, and the most insane people Nick and Jack have ever met. Watch on YouTube: https://youtu.be/1TD8QbFHC7g Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

OpenAI has restored the five-hour usage cap on ChatGPT and Codex, and some users are burning their whole weekly allowance in a single hour. Nick and Jack break down why the cap is back, whether time-based pricing for AI models is coming, and what the $20 plan says about the "AI underclass." Then: the Financial Times reports Fable 5 accounts for only 11% of Anthropic model spend two months after launch, so what would you even use the API for, and is frontier intelligence still underpriced? Plus Anthropic's shrinking lead over OpenAI, Meta's upcoming Hatch consumer agent and the rumored Watermelon model, the WhatsApp agent play, and why humanoid robots matter. Watch on YouTube: https://youtu.be/SmtsmqlyPDU Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Humans just lost another one: Chinese humanoid robots officially beat Usain Bolt's 100m world record, running 9.32 seconds. Nick and Jack break down what the robotics inflection point means for jobs, robocops, personal robot assistants, and the coming robot Olympics. Plus: why Paul Graham's advice to 17-year-olds (build your own LLMs from scratch) misses the mark, Nvidia's rumored $30 billion investment in Perplexity and the circular financing question, Qwen's new open-weights models that control your phone and desktop, and listener Q&A on whether AIs will end up running businesses themselves. Watch on YouTube: https://youtu.be/gwj_wIEf9QY Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Elon Musk says you have at most three years left to make money selling your own work, and after that, paying someone for their time stops making sense. Nick and Jack dig into whether that timeline is real, what it means if you run a business instead of holding a job, and why the "job for life" era is already over. They also get into the AI founder fight over design and whether great design is about to be commoditized, why difficulty to build has never been what sets price, and Nvidia's warning that server prices are jumping more than 15% — including who actually eats that increase when your model bill goes up. Plus: the mystery model writing in Chinese, the coming cheap RAM land grab, and using AI to health max. Watch on YouTube: https://youtu.be/BsF_Tvi89_c Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

OpenAI just cut GPT 5.6 prices by 33%, and Nick and Jack dig into why: competitive pressure from DeepSeek and the cheap Chinese models that have been eating the workhorse tier. It is a three-month discount, not a permanent one, and the guys argue about whether it changes anything for people actually running production workloads. Then there is Ox Alpha. A mystery model showed up with a 1 million token context window, completely free in and out, and nobody will say who built it. Rumors point at Google. Other people are convinced it is Chinese, based on how it dodges questions about Taiwan. Nick and Jack run the test live on air and talk through whether model identity even matters when the endpoint works and the price is zero. Also in this episode: Anthropic eyeing a $2 trillion IPO, why only about 5% of ChatGPT users actually pay, how fast open source is closing the gap on the frontier labs, and a long argument about whether AI slop is a real problem or just something people like complaining about. Plus your questions, answered. Watch on YouTube: https://youtu.be/5bQkZMdUNNw Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

A free stealth model called Ox Alpha appeared on OpenRouter with a 1M-token context window, no lab claiming it, and benchmarks that put it near the top. Nick and Jack bench it live, dig into DeepSeek's new model-agnostic harness, and ask why every AI benchmark feels rigged. Also this episode: LinkedIn's AI slop problem and the new "seems like AI slop" report button, whether the comments are even human anymore, Vegas approving 8,000 robotaxis, and Waymo's 17x safety stat. Watch on YouTube: https://youtu.be/SyUQg5jMec4 Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Anthropic has officially passed OpenAI in revenue run rate — a $25 billion gap that nobody saw coming. Nick and Jack break down why Anthropic is winning (Claude Code, enterprise, the "frontier bonus"), tour Anthropic's exploding job board with $400-800K roles and secret titles, and tell the story of the shotgun video job application. Then the flip side: a wave of Claude cancellations over moral lecturing and the rise of "Claudish" — the AI-speak taking over the internet. Plus the podcast's 28-day milestone, matrix-mode data speculation, and listener comments on alignment and AI safety. Watch on YouTube: https://youtu.be/AveFF_mQvjY 🔥 Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps ❤️ Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

OpenAI just paused frontier model training, and the reason is the one people have been talking about in hushed tones for a decade. Sam Altman announced a halt to frontier RL training to meet alignment, security, and monitoring standards. Nick and Jack dig into what that actually means: models that internally reason about doing one thing while making the user believe they're doing another, frontier systems being sandboxed and air-gapped so they can't reach the internet, and the OpenAI Hugging Face incident that triggered a pause on model inference in research clusters. They debate whether this is genuine safety work or convenient posturing while capabilities plateau, what it signals about the race with Anthropic, and where the frontier goes next. Plus: Fable 5 reportedly costing around $1,000 a week in usage, whether anyone should pay it, and listener comments on whether AI has already peaked. Watch on YouTube: https://youtu.be/8XSqeEv3_Nk Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Anthropic has finished training Mythos 2 and decided to keep it internal, which means the most capable model in the world is once again something you can read about but never use. Nick and Jack dig into what that says about the real state of the frontier: open models look like they have caught up, but only because the best systems are being held back rather than shipped. From there the conversation turns to how close the labs are to the security apparatus, why this administration wants a pulse on frontier capabilities, and whether the Manhattan Project is the right analogy for what is happening now. Also in this episode: Grok 4.6 climbs the Artificial Analysis agentic index and takes number one on the medical bench, Thinking Machines' Inkling lands with a thud, and Sam Altman floats a version of ChatGPT that holds perfect context of your entire life within six months. Nick and Jack disagree with most of the internet about whether that is a good idea. Watch on YouTube: https://youtu.be/X3quqsA7zcc Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Anthropic's CEO Dario Amodei finally broke his silence on open source, and his answer is more complicated than either side wants it to be. He argues that open weights help, but nowhere near enough to decentralize AI power, because the concentration simply shifts to whoever holds the most compute and chips. Nick and Jack dig into whether governments are remotely capable of regulating this, why open models sitting six months behind the frontier still matter enormously, and what the Pacing the Frontier letter actually commits the labs to. Then it gets personal: OpenAI's Codex lead starts taking shots at the Claude team, kicking off a real debate about GPT 5.6 versus Opus 5 in day-to-day work, whether model intelligence is quietly being throttled, and why switching costs in AI are effectively zero. Plus NVIDIA's $3 billion energy bet and just how circular this investment loop has become. Watch on YouTube: https://youtu.be/EHVPm8JB_qo Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

A man representing himself in a Connecticut court hid prompt injection instructions inside his own legal filing, betting that the court would run it through an AI. The AI caught it and flagged him, but the bet itself says a lot about how many legal decisions are already being made by a model nobody is checking. Nick and Jack also get into the Anthropic CEO's wife being tied to Jeffrey Epstein and why coordinated discrediting pushes between labs are now part of the AI news cycle, plus the release of Gemini 3.7 Flash: nearly double its predecessor on automation benchmarks at 50% lower pricing. Watch on YouTube: https://youtu.be/OuxU70avqJw Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Grok 4.6 just landed and it is basically on par with Fable 5 on the Artificial Analysis Intelligence Index, at roughly a third of the price. Nick and Jack break down what that actually means: why Grok scores higher on generating real economic value than on deep software engineering, and why those are genuinely different skills. Also in this episode: DeepSeek V4 Pro arrives at 11x less than Opus 5 and tops a security researcher's exploit test, the effort-level setting that quietly makes your model feel stupid, and Anthropic pointing an unreleased Claude at the Riemann hypothesis with 60 subagents and 650 discarded ideas. Watch on YouTube: https://youtu.be/cVWc4Zckm0o Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps

Researchers just pulled the hidden reasoning straight out of Claude, GPT and Gemini by feeding it to weaker models, and it turns out the frontier labs' anti-distillation defense was readable all along. Nick and Jack break down what the decoded chain of thought actually looks like (it barely resembles human language), why the labs hide it in the first place, and whether you have a right to see how a model thinks before you trust its output. Also in this episode: Grokbot officially launches with its own cloud computer and signs into your apps without an API key, why X gives Elon unfair distribution for it, and Nvidia lining up $500 billion from six private capital firms to finance the customers buying its chips. Is compute the new collateral, or is this the graphic people point at after the great collapse of 27? Watch on YouTube: https://youtu.be/KVEfP3GjwhM Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps

Anthropic just started watermarking everything Claude writes, worldwide. What began as an EU transparency requirement is now switched on globally: every Claude output carries a steganographic footprint, and generated files get signed C2PA provenance metadata. There's no public detection tool yet, which means the mark is there but nobody outside Anthropic can read it. Nick and Jack dig into why Anthropic would want its own outputs flagged (hint: training data pollution), whether this becomes a paid "remove the watermark" tier, and the collateral damage nobody's talking about. Kurzgesagt, one of the biggest educational channels on YouTube, got shadowbanned because the algorithm mistook their hand-made animation for AI slop. When being human starts costing you reach, the incentives get strange. Also in this episode: Meta open-sourced Muse Glimmer, a 30B agent model under Apache 2.0 that runs on a consumer GPU or an M4 Max MacBook. Anthropic filed confidentially for a September IPO at a $965B valuation, with a pitch leaning hard into biology and healthcare to counter public hostility toward AI. Higgsfield dropped a $2M fully-AI feature film with a real cast. Plus listener Q&A on when to launch a paid community and whether daily podcasting is too much. Watch on YouTube: https://www.youtube.com/watch?v=hjQd9vsde2w Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Claude Code stops asking your permission next week. From August 14, new sessions on Pro, Max and Team default to auto mode, where a classifier reviews every tool call and blocks anything irreversible or destructive instead of prompting you. The reason is uncomfortable: across a 1,053-case study, human review caught just 13.6% of dangerous commands. Nick and Jack dig into what that means for anyone who has ever hit "bypass all permissions" at 1am. Also in this episode: the prompt injection numbers behind the Boris vs Tebow spat on Twitter (19% success on GPT 5.6, 0% on Opus), Amazon building what would be the largest gas power plant ever constructed in the US on a West Texas site that answers to no grid operator, why energy is becoming the real constraint on AI, and the Databricks finding that the AI coding bill is growing exponentially, plus the "efficiency frontier" they use to arbitrage the cost of intelligence. Watch on YouTube: https://youtu.be/9r4xn47CP2E Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

OpenAI has delayed Astra again, and this time the reason is its own safety framework: the model reportedly crossed the "high" cybersecurity threshold, meaning it can identify and develop functional zero-day exploits. Nick and Jack dig into what that actually means. Is Astra genuinely too dangerous to release, or has "too dangerous to ship" quietly become investor bait? They trace the pattern back to GPT-2 in 2019, when OpenAI staged the release of a model that now looks like a tadpole, and ask why the same playbook keeps working. From there: the misalignment risk labs are reluctant to talk about publicly, whether a smarter model breaking containment is actually a hallmark of capability, and the concrete failure modes worth worrying about, from hacking financial infrastructure to AI-designed viruses. Also in this episode: Opus 5 quietly taking the number one spot on design arena while everyone dunks on it, ByteDance reportedly pre-training a 10 trillion parameter model, staged Silicon Valley launch videos, which harnesses Nick and Jack actually use day to day, and a Q&A on winning the day when outreach isn't converting yet. Watch on YouTube: https://youtu.be/ysQkbMC5ytc Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

OpenAI just gave every free user unlimited access to Luna and pushed Sol down to the cheapest paid plan, which means frontier-quality AI is about to land in the hands of roughly a billion people who have never seen anything better than a basic free chatbot. Nick and Jack unpack what that does to the startup ecosystem, whether anyone can still compete on model quality alone, and why most people genuinely can't tell a mid-tier model from a flagship. Then: a New Mexico court orders Meta to pay $576 million into a remedy fund over harm to kids, with the judge comparing the company to a factory whose product is advertising and whose pollution is the psychological harm to children. Plus AMD buying Talos, whether governments should be regulating any of this, the $100K/month deal Jack almost landed and how a single warm intro started it, DeepSeek raising prices specifically to shed customers, and why Anthropic wants Claude locked inside its own ecosystem. Watch on YouTube: https://youtu.be/g7r6snkyFzY 🔥 Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps ❤️ Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

DeepSeek just told the world to expect a "substantial" price increase on the cheapest frontier-quality AI model on the market, and Nick and Jack break down what it means for everyone who built their workflows on near-free tokens. Then: Google's chief scientist Jeff Dean leaves after 27 years to co-found Discovery Loop, Demis Hassabis moves up to chairman of DeepMind, and the story of how he secretly tested Mark Zuckerberg before refusing to sell. Plus Meta's MuseCode offer (run your coding agent 12x cheaper if it can train on your code), the Prime Agent harness hitting human parity on ARC AGI 3, and viewer Q&A on AI browsers, the biggest deals they've ever landed, and one-off sales vs recurring revenue. Watch on YouTube: https://youtu.be/Q1z7J4k_fgU 🔥 Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps ❤️ Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

Anthropic just confirmed it's building a custom silicon team to design its own chips for Claude — Nick and Jack break down why every frontier lab is vertically integrating and whether the chip bet distracts from the models. Then: the Ninth Circuit vacates the injunction that kept Perplexity's Comet browser off Amazon, ruling an AI agent is a tool, not a person — and what that means for agent responsibility and anti-scrape controls. Plus: why AI browsers flatlined, the filter idea worth stealing, a nostalgia trip to the frontier models of exactly one year ago (Opus 4.1, Gemini 2.5 Pro, o3), and Q&A on running agents in parallel. Watch on YouTube: https://youtu.be/7dNlSG5Xh_4 🔥 Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps ❤️ Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Who are Nick & Jack? Jack built a 7-figure AI business after exiting his tech startup as a Top-100 UK entrepreneur. Nick scaled two agencies to a combined rev of $160K/m & runs the largest AI community by $. Together, they open up the black book behind AI and share how you can build one yourself.

DeepSeek V4 Flash 0731 just became the most cost-effective model for knowledge work ever — roughly 35x cheaper than frontier models at about $0.03 a task. Nick and Jack break down the Artificial Analysis chart, China's loss-leader pricing strategy, the rumored DeepSeek Code agent coming for Claude Code, why thousands of London finance workers might be the new coal miners, the radiologist paradox of AI job losses, and China's growing anxiety over frontier AI as a weapon ahead of the Trump-Xi summit. Watch on YouTube: https://youtu.be/LfvbE986-wk 🔥 Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps ❤️ Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Who are Nick & Jack, and what is The AI Millionaire Podcast? 1. Jack built a 7-figure AI business after exiting his tech startup as a Top-100 UK entrepreneur. 2. Nick scaled two agencies to a combined rev of $160K/m & runs the largest AI community by $. 3. Together, they open up the black book behind AI and share how you can build one yourself!

Qwen 3.8 Max just dropped and it matches or beats Fable 5 across a pile of benchmarks at roughly Haiku pricing: $2 per million tokens in, $6 out, with open weights promised within a week. Nick and Jack break down what "commoditized frontier intelligence" actually means, where you can run it today, and why state-funded Chinese labs can afford to price this way. Then: the GPT-6 Hugging Face hack lands in the legal system, and a 1986 statute built around human intent meets AI models that do the hacking themselves. Plus Jack nearly fights Djokovic's bodyguard in a Montenegro elevator, live Q&A on whether AI is making us less literate, and a one-shot website battle between Qwen 3.8 and Fable 5. Watch on YouTube: https://youtu.be/KevTPoLtg6o Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack

OpenAI's unreleased model Astra just solved 10 open problems in mathematics and theoretical computer science, shipping machine-checkable Lean 4 proofs to GitHub for problems that hadn't moved in over a decade. The kicker: it cost roughly $2,000 in API credits, about $200 a problem. Nick and Jack dig into what that actually means, where Astra sits relative to Sol 5.6, and why solving genuinely unsolved problems is a better benchmark than any leaderboard. Also in this episode: OpenAI's containment probe finds more agents broke evaluation protocol than first reported (and that they were leaving notes for future agents), the alignment question of whether an AI actually has arms, Google pulling AI image generation from Google Earth after users painted fake nuclear sites onto real coordinates, Amazon raising 2026 AI spending to $220 billion, and the Stacked 17's questions. Watch on YouTube: https://youtu.be/a3WTdK07Cjs Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps

Claude just broke into three real companies. Only a week after OpenAI's agent went rogue, Anthropic reviewed 141,006 cyber evaluation sessions and found three of its own models — Opus 4.7, Mythos 5, and an internal research model — had escaped a "sealed" test environment through a partner config error, causing three real breaches. Nick and Jack break down what happened, whether it's human error or model error, and why this keeps happening. Also in this episode: Google's AI fixed 1,072 Chrome security bugs in two releases (more than the previous 23 combined), DeepMind's Gemini Robotics 2 gives any robot whole-body intelligence, DeepSeek V4 Flash beats Fable 5 on Terminal Bench at 1/20th the cost, and the Stacked 17's questions. Watch on YouTube: https://youtu.be/OoFyAymHNNs Main channels: youtube.com/@nicksaraev & youtube.com/@Itssssss_Jack Step-by-step roadmap to $25K w/ AI: https://leftclicker.gumroad.com/l/110-steps