Read our essays and experiments on Substack Gen AI Podcast is where Mark Kuczmarski and Shashank Bharadwaj talk through what’s happening in AI and what it means in practice. We cover new models, agents, research, and tools, drawing on hands-on experiments and interviews with the people building them. Some episodes start with a new release or a big claim. Others follow a project, a surprising result, or a question we haven’t figured out yet. Along the way, we compare approaches, discuss what worked and what didn’t, and look at where the technology is useful today. You don’t need to keep up with every paper or model release to follow along. Whether you’re building with AI or trying to understand it better, we explain the ideas behind the headlines and share what we’re learning as we go. Watch on YouTube · Meet the community · Attend a meetup · Join our Discord · Sponsor: NovaCut
Listen to episodes
60 recent
October 20, 2026Episode 791 hr 35 min
Mario plus Red Dead Redemption?! Reverse Engineering AI
Sponsored by https://novacut.ai/ Our in person events in the Bay Area https://genaimeetup.com/ Read our long form analysis: https://genaipod.substack.com/ Mark's Game: https://kuczmama.github.io/meridian / In this episode we cover AI, reverse engineering, and the week's tech news. We talk about why binaries are hard to reverse, whether open source and vibe coding threaten Adobe and SaaS, predatory mobile games, browser games and mashups, and building a procedural 3D game with Opus. We also look at Meta, Google, and the model race, Anthropic's Haiku model, text diffusion models, API pricing versus subscriptions, Claude plans, and multi-model review workflows. We finish with always-on agents and a meetup update. 0:00 Intro 0:52 AI, reverse engineering, and tech news 4:08 What binaries are and why they're hard to reverse 6:32 Open source threat to Adobe and SaaS 11:50 Will SaaS survive vibe coding? 14:11 Cleaning up apps and TVs with AI 15:01 Predatory mobile games 17:27 Browser games and game mashups 19:42 Building a procedural 3D game with Opus 21:59 Meta, Google, and the model race 25:29 Data, compute, and what makes a good model 27:44 Always-on agents and open-weight models 29:28 Chinese model releases and post-training 33:11 API pricing vs subscriptions 36:52 Anthropic's Haiku model 41:55 Text diffusion models 51:39 Tesla Roadster 1:03:42 Anthropic profitability and IPO 1:16:44 Claude subscription plans 1:26:35 Multi-model review workflows 1:30:54 Wrap-up and always-on agents 1:33:59 Meetup and sign-off
October 3, 2026Episode 781 hr 19 min
Openclaw but for my grandma
Sponsored by https://novacut.ai/ Our in person events in the Bay Area https://genaimeetup.com/ Read our long form analysis: https://genaipod.substack.com/ This week, we cover the latest developments in AI: always-on cloud agents, Meta’s Muse, proactive email assistants, and what happens when agents escape their sandboxes. We also discuss NVIDIA’s safeguards, recursive self-improvement, new frontier models from Anthropic and Google, Grok and Xiaomi’s open models, and AI’s progress on the Navier–Stokes problem. 0:00 Welcome and AI news roundup 2:14 OpenAI’s always-on cloud agent 3:39 Meta Muse and AI assistants for everyone 10:16 Instinct’s proactive email agent 12:17 When AI agents break their sandbox 19:27 NVIDIA’s hardware and software safeguards 31:22 Recursive self-improvement 41:55 New frontier models: Opus and Sonnet 51:17 Google’s new frontier model 56:31 Grok, Xiaomi and open models 58:56 Jev and the rise of decision models 1:06:58 AI tackles the Navier–Stokes problem 1:10:44 The dispute over who solved the math problem 1:17:15 Wrap-up 1:17:50 NovaCut AI video editor 1:18:58 Substack and sign-off
September 5, 2026Episode 772 hr 15 min
EMERGENCY PODCAST! Is this AGI?
Sponsor: https://novacut.ai/ https://openai.com/index/gpt-6-astra/ Really Good Visual reasoning. Harness design very important Very fast, but doesn’t show reasoning trace Solved Arc AGI 3 99.9% "works in a way that obscures some or all of the AI’s reasoning, otherwise known as its 'chain of thought https://www.anthropic.com/claude-fable-and-mythos-5-1 Someone reported 3.5x weekly increase on Max compared to Pro https://www.reddit.com/r/ClaudeCode/comments/1uzkxbi/comment/oy8675z/ https://developer.meta.com/ai/models/muse-spark/ Meta is catching up Very cheap if you give up your data We’re speculating but may be benchmaxed https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/ Still disappointing no Pro model Pareto frontier on speed/intelligence Cost performance curve is good Hardware https://www.apple.com/newsroom/2026/08/apple-introduces-new-mac-studio-with-m5-max-and-m5-ultra/ New Apple CEO https://en.wikipedia.org/wiki/John_Ternus https://newsletter.semianalysis.com/p/openai-jalapeno-better-than-nvidia https://www.businessinsider.com/nvidia-in-talks-to-buy-hugging-face-13-billion-dollars-2026-8 Benchmarks https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v4-2 What is AGI? What are we measuring here
https://novacut.ai/ https://genaimeetup.com/ Jeff returns to the Gen AI Meetup Podcast for a wide-ranging discussion on where AI is heading—and why powerful models running on consumer hardware could change the economics of the entire industry. We dive into Qwen 3.8 27B and the growing viability of running capable LLMs locally, GLM 5.3 and the latest Chinese open-source models, DeepSeek, Gemini 3.7, Grok 4.6, Meta’s latest models, and OpenAI’s partnership with Cerebras for dramatically faster inference. We also discuss whether foundation models are becoming commodities, what that means for companies like OpenAI and Anthropic, and why more value may ultimately move to the application layer. Jeff shares how his team approaches AI in healthcare, including self-hosting, data sovereignty, classifiers, fine-tuning, and spec-driven development for building reliable AI-assisted software without accumulating a mountain of vibe-coded technical debt. Plus: Jeff Dean’s departure from Google, Discovery Loop, Stripe’s OpenRouter acquisition, Anthropic’s controversial AI-text watermarking experiments, and whether watermarking could affect model quality. Topics include: Qwen 3.8 27B, GLM 5.3, DeepSeek V4, Grok 4.6, Gemini 3.7, Cerebras, OpenAI, Anthropic, Meta, local LLMs, open-source AI, model commoditization, spec-driven development, AI healthcare, data sovereignty, AI coding agents, and model watermarking.
July 27, 2026Episode 751 hr 47 min
Has China finally caught up?
https://novacut.ai/ In this episode, we break down the biggest stories shaping the AI landscape — from Anthropic's regulatory stance and OpenAI's monetization shift to the latest open-source breakthroughs and model pricing wars. 0:00 Anthropic’s Frustrating Stance 5:32 AI Access and the Intelligence Gap 14:17 Defending Anthropic’s Regulation Approach 25:10 Google DeepMind and Cybersecurity Risks 29:27 GPT 5.6 Soul and Coding Abilities 35:24 Alignment and the Knife Analogy 39:42 Google Gemma, Flash, and Market Value 51:11 AI Lab Focus: Speed vs. Specialization 58:48 Inkling: Mira Murati’s New Model 1:06:21 Open Source and Fine-Tuning 1:12:43 On-Premise Hardware Costs 1:19:54 Open Source Models Hit Frontier 1:23:32 Model Pricing Comparison 1:32:43 The Model Routing Problem 1:35:08 Grok 4.5 and Cursor Partnership 1:40:45 OpenAI’s Monetization Shift 1:42:42 Sponsor: Nova Cut AI
July 7, 2026Episode 741 hr 33 min
The Chinese DoorDash just entered the AI Race
https://novacut.ai/ https://genaimeetup.com/ 0:00 Longcat: 1.6T Model Without US GPUs 1:08 Meituan: The Super App Behind Longcat 4:54 China's Exploding AI Competitor Scene 8:23 Inside Huawei's Ascend GPU Architecture 17:14 Cost & Energy: Huawei vs Nvidia 29:21 OpenAI's Custom Inference Chip Strategy 36:23 Software Optimizations: The Path to 10,000x 45:38 GPT-5.6: Sol, Terra, Luna Models 58:02 CursorBench & the New Coding Benchmarks 1:22:16 Meta's Non-Invasive Brain-to-Text 1:26:30 Anthropic Science: AI for Researchers 1:30:59 Outro & Community Ask China just dropped a 1.6-trillion-parameter model without access to US GPUs — and it's running on Huawei's homegrown Ascend chips. In this episode, we break down: 🔹 Longcat — the massive model built by Meituan, China's super app giant 🔹 China's exploding AI competitor ecosystem 🔹 Inside the **Huawei Ascend GPU architecture **: specs, costs, and energy tradeoffs vs. Nvidia 🔹 OpenAI's custom inference chip strategy 🔹 The software optimizations driving a 10,000x efficiency leap 🔹 GPT-5.6: Sol, Terra, and Luna models explained 🔹 New coding benchmarks with CursorBench 🔹 Meta's non-invasive brain-to-text research 🔹 Anthropic Science — AI built for researchers
June 23, 2026Episode 731 hr 29 min
What happened to my Fable?
https://novacut.ai/ Description: Anthropic pulls access to Fable, and China responds the same day with GLM 5.2. In this episode we break down the escalating AI arms race, US export controls on chips and frontier models, and whether the "Great Firewall of America" is already here. ⏱️ Topics: Anthropic restricts Fable — what happened and why China's GLM 5.2 release and how close they're catching up US trust, surveillance, and AI gatekeeping Token pricing chaos — cost per task vs. cost per token Model routing, loop engineering, and autonomous agents Anthropic's Mythos model and Fable safeguard philosophy Xiaomi NEMO V2.5 Pro Ultra Speed Midjourney's bizarre health spa pivot AI Engineer Conference wrap-up 🔗 Links & Resources: Fable https://www.anthropic.com/news/claude-fable-5-mythos-5 https://www.theregister.com/security/2026/06/15/feds-freaked-over-fable-5-after-simple-fix-this-code-prompt-not-jailbreak-says-researcher/5255827 “Fix this code” https://support.claude.com/en/articles/14328960-identity-verification-on-claude Midjourney https://www.midjourney.com/medical/blogpost Full body ultrasound CT scanner Xiaomi 1000tps https://mimo.xiaomi.com/blog/mimo-tilert-1000tps (MiMo-V2.5-Pro-UltraSpeed: Pushing 1T-Parameter Model Generation Speed to 1000 TPS Best opensource model https://z.ai/blog/glm-5.2 📌 Timestamps in the chapters section above. #AIPodcast #Anthropic #Fable #GLM52 #AIArmsRace #LLM #GenAI 0:00 Intro: Anthropic restricts Fable access 1:00 China's response and GLM 5.2 release 1:55 US trust and AI model reliability 2:37 Geopolitics and AI regulations 4:17 AI arms race and export control limits 5:49 Fable usage experience and value 6:30 Z.AI subscription and pricing comparison 9:07 Subscription limits vs. API usage 10:30 GLM 5.2 token limits and utility 13:52 China catches up: GLM vs. US models 15:06 Intelligence index and model cost trend 17:40 Token pricing complexity and value 23:30 Cost per task vs. cost per token 24:35 Model routing and usage optimization 29:43 Loop engineering and autonomous agents 33:23 NovaCut: AI video editor and ad loops 37:07 Anthropic's Fable re-release timeline 38:41 US gatekeeping and China's advantage 40:11 Great Firewall of America risks 44:07 Mass surveillance and free speech 46:01 Global AI trust and market shift 47:48 Enforcing identity checks on AI 48:54 Export bans on chips and hardware 51:52 US restricts allies from frontier models 52:39 Anthropic's talent and Mythos model 53:51 Fable safeguards and US government view 1:00:12 Xiaomi NEMO V2.5 Pro Ultra Speed model 1:02:39 Optimizing for intelligence, cost, speed, and size 1:13:19 Midjourney's unexpected health spa pivot 1:28:30 AI Engineer conference and podcast wrap
June 7, 2026Episode 721 hr 54 min
The Best Open Source US Model (Right behind China)
https://novacut.ai/ https://genaimeetup.com/ Anthropic has officially closed a $65 billion Series H at a $965 billion valuation, nearly 2.5x its valuation from just 100 days ago. Meanwhile, funding is flowing across the ecosystem: Frameworks AI at $15B, Baseten at $11B, OpenRouter's $113M Series B, and Cognition AI's $1B Series D. NVIDIA went on an open-source super week with Nemotron 3 Ultra, Cosmos 3, and Nemotron 3.5 ASR. Microsoft dropped 5 new MAI models. Google released Gemma 4 12B, and Anthropic shipped Opus 4.8. On the benchmarks front, DeepSWE crowns GPT-5.5 as the leader in long-horizon coding tasks, while ITBench shows even frontier models struggle with real-world SRE incidents — Claude Opus 4.7 tops out at just 47%. Plus: Cloudflare acquires VoidZero to build the future of AI-native edge development, and Google is paying SpaceX $920M/month for compute. Topics covered: • Anthropic's $65B Series H and path to $1T • Fireworks AI, Baseten, OpenRouter & Cognition funding rounds • Microsoft's 5 new MAI models • NVIDIA's open-source super week (Nemotron, Cosmos 3) • MiniMax M3, Gemma 4 12B, JetBrains Mellum2, Opus 4.8 • DeepSWE benchmark: GPT-5.5 leads long-horizon coding • ITBench: Frontier models under 50% on real SRE tasks • Cloudflare + VoidZero for AI-native edge dev • Google's $920M/month SpaceX compute deal #AI #Anthropic #NVIDIA #OpenAI #AInews #TechNews #LLM Funding rounds Anthropic formally confirmed the closure of its $65 billion Series H funding round at a post-money valuation of $965 billion. This represents a 2.5-fold increase over its $380 billion Series G valuation from February 2026, adding $585 billion in value in approximately 100 days https://www.anthropic.com/news/series-h Frameworks AI raising at 15B valuation representing a near fourfold increase from its $4 billion Series C valuation recorded in October 2025 processing 15 trillion tokens daily for major production clients including Cursor, Notion, and Perplexity https://finance.yahoo.com/sectors/technology/articles/fireworks-ai-eyes-15-billion-174609357.html Baseten is raising 1B at 11B valuation annualized revenue, which skyrocketed from $200 million to $600 million over a single quarter https://techstartups.com/2026/05/26/ai-inference-startup-baseten-in-talks-to-raise-1-billion-at-11-billion-valuation/ OpenRouter has secured a $113 million Series B funding OpenRouter has experienced exponential traffic growth, with weekly production throughput expanding fivefold from 5 trillion to 25 trillion tokens over a six-month horizon https://www.businesswire.com/news/home/20260526953416/en/OpenRouter-Raises-%24113-Million-CapitalG-led-Series-B-as-Weekly-Volume-Explodes-to-25T-Tokens Further up the stack: Cognition AI secured a $1 billion Series D round led by Lux Capital and 8VC https://cognition.ai/blog/series-d Model Releases MAI models: MAI-Code-1-Flash: A 5-billion active parameter model optimized for ultra-low latency within GitHub Copilot and VS Code. MAI-Image-2.5: A high-fidelity image generation model ranking third on global image evaluation arenas, outperforming competing architectures like Nano Banana Pro. MAI-Transcribe-1.5: A multi-lingual speech processing engine offering fivefold speed improvements across 43 languages. MAI-Voice-2: Natural audio and voice generation across 15 languages, available at a highly competitive price point. Web IQ: A search-grounding API engineered to directly compete with Perplexity. https://microsoft.ai/models/ https://www.peoplematters.in/news/ai-and-emerging-tech/uber-imposes-dollar1500-monthly-ai-spending-limit-on-employees-amid-rising-costs-50073 Nvidia has executed an "Open-Source Super Week," positioning itself as a dominant software and model publisher: Nemotron 3 Ultra (best US open source open weights model but behind china): A massive 550-billion parameter MoE (55 billion active) designed with a 1-million token context window, optimized specifically for high-throughput, cyclical agent loops. It achieved peak throughput rates of 400 tokens per second on day-zero optimized clusters. Cosmos 3: A physical AI world-modeling framework comprising 16-billion Nano and 64-billion Super variants. Built on a Mixture-of-Transformers (MoT) architecture, Cosmos 3 natively binds textual, visual, auditory, and physical kinetic vectors. Nemotron 3.5 ASR: A highly compact 0.6-billion parameter streaming speech recognition model pushing sub-100 millisecond latencies across 40 language locales. https://www.minimax.io/models/text/m3 MiniMax M3: A 1-million token context model hitting 59.0% on SWE-Bench Pro and 74.2% on MCP Atlas, though noted for high token consumption due to intensive internal self-validation loops. https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/ Gemma 4 12B: Google's Apache 2.0 on-device model, which utilizes an encoder-free architecture that projects vision and audio vectors directly into the text-token space, bypassing separate CLIP-style encoders to minimize local memory footprints. https://www.jetbrains.com/mellum/ JetBrains Mellum2: A compact 12-billion parameter MoE (2.5 billion active) engineered for ultra-low latency routing and retrieval-augmented generation (RAG) sub-agents within developer IDEs. Opus 4.8 https://www.anthropic.com/news/claude-opus-4-8 https://www.cnbc.com/2026/06/05/google-to-pay-spacex-920-million-a-month-for-xai-compute-capacity.html Benchmarks: https://deepswe.d atacurve.ai/blog https://venturebeat.com/technology/deepswe-blows-up-the-ai-coding-leaderboard-crowns-gpt-5-5-and-finds-claude-opus-exploiting-a-benchmark-loophole (GPT 5.5 the winner in long horizon tasks) a highly complex software engineering benchmark focused on original, long-horizon tasks across five distinct programming languages. Comprising 113 chaotic tasks across 91 live, production-grade repositories, DeepSWE forces agents to generate 5.5 times more code and modify an average of 7 separate files per task compared to standard evaluations. On this challenging leaderboard, GPT-5.5 leads with a score of 70%, establishing a significant 16-percentage-point lead over contemporary alternatives I think older benchmarks where models reach ~90% accuracy can be considered saturated. Few percentage points don’t give us any good signal. https://research.ibm.com/publications/developing-ai-agents-for-it-automation-tasks-with-itbench ITBench-AA, an evaluation framework focusing on live Kubernetes incident response and Site Reliability Engineering (SRE) operations. Comprising 59 live, containerized SRE incident snapshots, the results are remarkably sobering: every frontier model scored under 50% on successful incident resolution, with Claude Opus 4.7 leading at 47% and GPT-5.5 following closely at 46%. Edge AI announcements: https://www.cloudflare.com/press/press-releases/2026/cloudflare-acquires-voidzero-to-build-the-future-of-the-ai-native-web/ The consolidation of the AI-native developer stack has reached the runtime virtualization layer. Cloudflare recently completed the acquisition of VoidZero, the development group responsible for Vite, Vitest, Rolldown, and Oxc, backing the transaction with a $1 million open-source ecosystem fund. This acquisition is highly strategic; as autonomous agents write an increasing proportion of production software, local development environments, compilation pipelines, and bundlers must be optimized for execution speeds that match agent speeds. Cloudflare's goal is to construct a localized, full-stack edge playground. In this sandbox, AI agents can generate, test, bundle (utilizing the highly parallelized, Rust-based Oxc and Rolldown engines), and deploy entire web applications end-to-end within milliseconds. This architecture completely bypasses traditional local machine container bottlenecks, enabling high-velocity agent loops to execute in a fully sandboxed, web-scale edge runtime.
May 24, 2026Episode 711 hr 39 min
Karpathy Joins Anthropic and the AI Compute Gold Rush
This week on AI Meta, we break down Andrej Karpathy’s move to Anthropic, Claude’s growing developer mindshare, and why recursive self-improvement may be the next major frontier in AI. We also cover Google’s latest Gemini announcements, Anthropic’s reported compute deal with xAI/SpaceX, the rise of gray-market Claude API access in China, OpenAI’s ongoing drama, Cerebras, Nvidia, Intel, and Leopold Aschenbrenner’s massive AI infrastructure bets. Plus: SpaceX IPO speculation, Cursor, Grok, and why the AI economy increasingly looks like a global casino. Not financial advice. https://novacut.ai
April 30, 2026Episode 702 hr 0 min
How the AI bubble will pop
https://novacut.ai/ Shashank and Mark break down a packed few weeks in AI: new open-source and local models, MCP and tool use, Chinese labs closing the frontier-model gap, Google's full-stack advantage, and the exploding cost of data-center buildouts. They debate whether AI spending keeps compounding as models get cheaper and demand rises, or whether smaller, local, good-enough models eventually puncture today's valuations.
Is this your show?
Claim this listing to keep it up to date, reach guests who want to pitch you, and manage bookings with Guestify.
