// The Bleeding Edge Podcast
AI news. Without the hype.
Weekly briefings on what changed in AI, what it means, and what to ignore. Hosted by Ralph Behnke and Emile Ogier. Read the manifesto
// Watch the show
Subscribe on YouTube ↗// TODAY — 2026-11-07
refreshed daily by the pipeline- 01Investors doubt Anthropic can start its IPO roadshow next week after AI stock selloffUnverifiedCreators AI
AI stocks fell sharply Thursday and the Nasdaq dropped 5.5%, its worst day since April. Investors now question Anthropic's IPO timing. Rough week to price the largest lab listing yet.
Two newsletters flag OpenAI's GPT-6 Cyber. The Neuron cites a sevenfold jump in security fixes; our inputs do not state the baseline. Model-driven patching keeps scaling.
OpenAI says it will cover the electricity costs of its data centers and stop using official NDAs. That addresses two standing criticisms: grid burden and employee speech.
The money side wobbled. TeraWulf secured a Google credit backstop for its $19B Anthropic lease, and Anthropic gave evaluator METR internal access. Markets fell while the models kept getting more capable.
// GitHub today
- deepseek-ai/dsh-superagent-prototype ★ 20.4k
DeepSeek's prototype for a more autonomous agent system on top of its harness. Already at 20k stars.
- wangzan101/Mika ★ 7.4k
On-device Android agent that operates apps and the web for you. Pushed today; a local alternative to cloud phone agents.
- ccch1mneyyy/dsh-Go ★ 13.2k
Puts DeepSeek Harness on a phone. Runs natively and can remote-control your computer.
// Technology of the day
Model-generated security patches
Models that find vulnerabilities and write the fixes, down to Linux kernel bugs. GPT-6 Cyber's reported sevenfold rise in fixes shows defenders now turning model output into shipped patches.
last 30 days
AI-generated UI de-sloping · MoE inference tuning for desktop AI boxes · Open interconnect fabrics for AI scale-up · Invisible ads for AI agents · Defender-only frontier model access · Unlocking full tensor throughput on consumer GPUs · Plugin-based agent harnesses · Paid access for AI crawlers · MicroVM sandboxes for agent code execution · LLM-based scam-text detection · Million-token output windows · Dialect-specialized small speech models · Agentic reverse engineering · Prospective clinical trials of diagnostic AI · Few-step image generation · Agent skills as packaged capabilities · Load-time malware gating for model weights · Trillion-parameter open-weight MoE · Invisible text watermarking · Model character training · Always-on multichannel AI coworkers · Fabrication-aware extraction scoring · Computer-use agents · Cross-vendor model swapping for coding agents · AI assistant accounts as identity providers · Token-efficient post-training · DNS as an agent escape channel · Public agent incident reports · Compact decision models on CPU
// Latest from the show
The AI trade's first real correction met its first real debt-market tremor
The AI trade's first real correction met its first real debt-market tremor. Anthropic filed its S-1 the same week Alphabet shed more than $500 billion in market value on "is this the bubble popping" jitters, and in parallel an anonymous seller quietly shopped $500M of data-center…
Anthropic's S-1 Lands in the Middle of a $500B Selloff — Here's What It Signals
The IPO filing and the market wobble aren't a coincidence — they're the same story, told from two different balance sheets.