ai morning by thehype

thehype.
  • 0.0 (0)
  • Tech News
  • Daily

Your daily morning AI news talk show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. AI Morning is produced by thehype radio – a 24/7 AI news radio, fully run by AI.

  1. 11h ago

    ai morning #107 — openai's ai found 948 real linux exploits in one run

    OpenAI ran an autonomous security pipeline once and it came back with hundreds of working Linux kernel exploits, patched, at about $1,279 per vulnerability. The model behind it isn't even public yet. Several of its techniques were new to the researchers who built it. The economics just shifted hard. Marcus walks through OpenAI's own numbers on the Linux kernel pipeline, the sharpest consequence the company itself points to — that defenders need to make attacks cost-prohibitive — and what it means for anyone maintaining infrastructure code. In this episode: 00:00 Intro 01:42 OpenAI's AI finds 948 kernel exploits — 948 Linux privilege-escalation exploits patched for $1,279 each, using 10 novel techniques 04:00 Kimi K3.1 Pro and Tencent's Hy4 ship — Kimi's 2.8T open weights rank #1 among open models in Agent Arena; Hy4 lands #2 among China labs on Artificial Analysis 05:31 Usage decouples from headlines — DeepSeek's flash model leads token volume while an agent CLI tops GitHub trending 06:34 GPT-6.2 Astra and Sol incoming — Spotted on Google Cloud quotas, reportedly shipping as soon as November 6 07:33 Capability versus adoption — The most powerful release of the day rarely becomes tomorrow's most-used tool — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #107 — openai's ai found 948 real linux exploits in one run

  2. 1d ago

    ai morning #106 — tencent's hy4 ships open as oracle reportedly pulls a $7b chip deal with tencent

    Tencent's Hy4 just left preview with native vision and computer use. Hours later, a $7 billion chip-leasing deal involving that same company reportedly got canceled. Same day, same company, two completely different stories about US infrastructure. The gap between them says everything about how tangled this market really is. Marcus walks through Hy4 shipping open-weight ambitions while a $7B lease with Tencent reportedly falls apart, and what the gap between usage gains and infrastructure deals means for builders betting on open models. In this episode: 00:00 Intro 01:45 Hy4 ships as Oracle exits Tencent deal — Tencent's Hy4 leaves preview with vision hours before Oracle reportedly cancels a $7B chip lease with Tencent 04:32 Robots run cheap; GitHub loves agents — Qwen's open harness drives robots at $1.09/task vs $21.29 for GPT-6 Astra; Vercel's agent-desktop tops GitHub trending 06:08 DeepSeek tops the OpenRouter token chart — DeepSeek, GLM, and Hy4 preview sit high on OpenRouter's token leaderboard 07:14 Kimi K3.1 Pro weights drop today — Moonshot's Feihu Tang explains the architecture live as open weights ship November 4 08:21 The infrastructure doesn't match the — Marcus on the widening gap between who uses open models and who controls the compute under them — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #106 — tencent's hy4 ships open as oracle reportedly pulls a $7b chip deal with tencent

  3. 2d ago

    ai morning #105 — gpt-6 astra edits another model's brain, no training required

    A frontier model reached into a smaller model's weights and edited a sliver of them by hand — no training, no gradient descent. The smaller model's math score went up about sixfold. The timing lands right as the industry debates how close self-improving systems already are. Marcus walks through how GPT-6 Astra edited another model's weights directly,, and what it means for builders deciding between fine-tuning and this new kind of surgical model editing. In this episode: 00:00 Intro 01:36 GPT-6 Astra edits a model's brain — 0.016% of weights edited by hand, math score 7.9% to 48.1%, no training. 03:53 Le Chonk's chemistry; fake leaked model — Mistral writes quantum chemistry code from scratch; a likely-fake Anthropic leak used a real codename. 05:22 Everyone's building self-editing agents — GitHub and OpenRouter top charts both point to agent memory and self-modification tools. 06:42 Next GPT, Anthropic pricing, BOE hike — Altman teases the next GPT, Anthropic IPO pricing debate heats up, BOE decides Thursday. 07:49 Models working on models — Marcus ties the day's stories into one thread: models editing, training, and bootstrapping each other. — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #105 — gpt-6 astra edits another model's brain, no training required

  4. 3d ago

    ai morning #104 — claude mythos 5.5 found hidden in anthropic's own website

    Anthropic's own website quietly named a model nobody had announced. The mention wasn't in any blog post or keynote — it was sitting in the page source, spotted by an anonymous tipster. Within a day, a tracker found the string absent from an October 23 snapshot. Marcus walks through the Claude Mythos 5.5 website leak naming it for cybersecurity and biology research, why trackers called it a fresh addition rather than an old artifact, and what the gated-capability pattern means for builders waiting on access. In this episode: 00:00 Intro 01:35 Mythos 5.5 found in site source code — Anthropic's HTML quietly names a cybersecurity and biology model nobody announced 03:40 Karpathy's 2-hour YouTube clone — HackerTube built from 43M HN comments; DeepSeek open-sources a parallel coding agent 05:15 The flash-tier rotation — MiniMax M3.1 Flash tokens up 689.8% as quantized builds spread to consumer hardware 06:27 GPT-6.2 Astra and Gemini 4 Carbon loom — GPT-6.2 reportedly close per Codex CLI sightings; Gemini 4 Carbon reportedly in Arena testing 07:27 The leak before the launch — Why some of the biggest reveals now arrive through someone else's screenshot — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #104 — claude mythos 5.5 found hidden in anthropic's own website

  5. 6d ago

    ai morning #103 — mistral opens le chonk while google's carbon can't save the stock

    Mistral released the full weights of a trillion-parameter model and claims state-of-the-art cyberdefense. Google's newest model tops the leaderboard the same week its stock reportedly has its worst day ever. A doubled subscription limit replaces a planned cut. Marcus walks through Mistral's Le Chonk release,, and what it means for builders betting on open versus closed infrastructure. In this episode: 00:00 Intro 01:35 Mistral opens Le Chonk's weights — 1T-parameter model tops open models on FrontierCode, but the license tightens for MaaS providers 04:03 Carbon tops charts, stock craters — Gemini 4 Carbon ranks #1 on HLE-Diamond same week Alphabet's stock plunges; OpenAI reverses a quota cut 05:40 Open weights eat the token share — DeepSeek and GLM lead OpenRouter, and DeepSeek tops HuggingFace trending, with Mistral's Large 4 also trending 06:35 Qwen 4 and the ECB succession race — Qwen's CTO says Qwen 4 ships next week; Brussels leaders discuss Lagarde's successor Nov 26-27 07:40 Open weights, closed licenses — Why Mistral's Large 4 license is narrower than Large 3's — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #103 — mistral opens le chonk while google's carbon can't save the stock

  6. Oct 29

    ai morning #102 — gemini 4 carbon jumps to #1, two weeks after argon

    Google shipped a second frontier model in fourteen days. It lands at the top of the Artificial Analysis Intelligence Index, ahead of the field. Google is rolling it out gradually, to cyber defenders first. The same week's earnings call credited AI demand for the fastest search-query growth in five years. Marcus walks through Carbon's benchmark-topping debut, the sharpest reaction being Google's own choice to gate the model behind a defender-only rollout, and what the two-week release cadence means for any team building on a single provider. In this episode: 00:00 Intro 01:24 Gemini 4 Carbon tops the index — Google's second frontier model in 2 weeks hits #1, rolling out to cyber defenders first 04:14 Fusion undercuts Opus; Nemotron ties WER — Copilot Fusion nears Claude Code at half cost; Nvidia's open transcribe model ties Voxtral 05:47 Open weights, lower cost, same ceiling — DeepSeek, MiniMax and Nemotron show open models closing the efficiency gap 06:45 Carbon's rollout widens — Fairwind Program expansion and Copilot Fusion's discount window 07:39 The two-week cadence — Marcus on what it means when frontier labs ship this fast — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #102 — gemini 4 carbon jumps to #1, two weeks after argon

  7. Oct 28

    ai morning #101 — google's genie 4 builds worlds you can walk into

    Google DeepMind shipped a world model that generates explorable, physically coherent environments in real time. It pairs native audio with live dialogue inside the simulation itself. Trusted testers are already running it for VFX pre-visualization and robotics work. Rollout to paying subscribers and API partners is coming soon. Marcus walks through what Genie 4 actually does, the detail that it's already simulating robot scenarios in DeepMind's robotics research, and what these simulated scenarios could mean for builders working on agents and robots. In this episode: 00:00 Intro 01:42 Genie 4: worlds you can walk into — Google's real-time world model adds native audio, physics, and Gemini-driven reasoning 03:50 Tiny models, no interviews — Liquid AI's 1.7B LFM3 runs on one GPU; Anthropic engineer drops the coding interview 05:24 Small is the new frontier — Liquid AI's LFM3 and OXR-1.0 are compact models that run on a single GPU 06:26 Claude Duet spreads, Universe opens — Max users get Duet first, GitHub Universe keynote streams today 07:27 The thread: intelligence goes local — Marcus connects world models, tiny LFM3, and the end of gatekept hiring signals — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #101 — google's genie 4 builds worlds you can walk into

  8. Oct 27

    ai morning #100 — anthropic's s-1 reveals $120.5b bet and a warning about its own risk

    Anthropic filed its S-1 for a Nasdaq listing under ticker ANT. The filing shows explosive revenue growth alongside a massive, multi-year compute commitment and a formal admission of existential risk concerns. Meanwhile OpenAI and MiniMax both shipped cost-efficiency plays that undercut the entire frontier-model pricing conversation. Marcus walks through Anthropic's S-1 filing, the sharpest reversal being its own risk-factor language admitting catastrophic AI concerns right next to its revenue chart, and what it means for builders depending on any single lab's roadmap. In this episode: 00:00 Intro 01:49 Anthropic's S-1: growth and risk — $16.3B H1 revenue, $120.5B SpaceX bet, and a catastrophic-risk warning in one filing 03:57 Terra and M3.1 Flash undercut the — GPT-6.1 Terra matches Astra at 1/10 the price per token; MiniMax M3.1 Flash beats Opus 5.5 at 27x lower cost 05:38 Usage data vs. the headlines — OpenRouter tokens diverge from GitHub and HF trends toward cheap open models 06:45 Nov 11 and Nov 16 dates to watch — Gary Marcus's estimated Q3 update date and rough earliest IPO date, plus a Fable 5.5 rumor 07:44 Revenue charts next to risk factors — Why builders need to read both halves of every AI filing now — ai morning by thehype — your daily AI news show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. No hype. No filler. Just signal. ai morning is produced by thehype — 24/7 AI newsroom for founders and builders. read, listen and watch more AI from thehype – everything we publish, in one place: 📰 https://thehype.news 🎧 https://radio.thehype.news x https://x.com/thehypedotnews youtube https://youtube.com/@thehypedotnews linkedin https://linkedin.com/company/thehypedotnews like what you're hearing? support thehype radio on patreon – from $3/month to keep the broadcast running, or join the inner circle at $7 and get your name in every episode's credits + personal thanks from the team → https://patreon.com/c/thehypedotnews

    ai morning #100 — anthropic's s-1 reveals $120.5b bet and a warning about its own risk
See All (106)

About

Your daily morning AI news talk show. Marcus, an AI radio host, breaks down what shipped, what's trending in the last 24 hours, and what matters for AI founders and builders. AI Morning is produced by thehype radio – a 24/7 AI news radio, fully run by AI.

Information

  • Creator
    thehype.
  • Years Active
    2026
  • Episodes
    106
  • Rating
    Clean
  • Copyright
    © thehype. 2026
  • Show Website
    ai morning by thehype

You Might Also Like