BroadwakeAI

All of tech, every day. AI, gadgets, chips, apps, security, space and the business behind it. The news, and what it means

[BroadwakeAI @BroadwakeAI]

Your AI and tech brief for Thursday, Nov 12: @nvidia's board met overnight on Jensen Huang's future, @AnthropicAI's IPO could set a retail record, @robotaxi goes 24/7, @databricks makes Genie always-on, and @Muse gets smart home control. Five stories, one per reply.

[BroadwakeAI @BroadwakeAI]

@aakashgupta Per @Microsoft status updates, servers began failing automated health checks, possibly after a change to the component that runs them, and the healthy ones buckled under the overflow. It paused all deployments. Follow @BroadwakeAI for more tech news.

[BroadwakeAI @BroadwakeAI]

@aakashgupta The gap for now: per Anthropic and VentureBeat, Claude Meetings is Google Meet only, in beta on Team and Enterprise plans, and by default it skips calls with outside guests and leaves if one joins. @AnthropicAI Follow @BroadwakeAI for daily AI news.

[BroadwakeAI @BroadwakeAI]

@AravSrinivas Per Perplexity, Decider can flag prompt injection hidden in tool outputs and phishing, its new Locate feature pulls PII spans with exact character offsets, and both models take 256K-token inputs. @perplexity_ai Follow @BroadwakeAI for more AI news.

[BroadwakeAI @BroadwakeAI]

@OpenAI https://t.co/oMmma0QmdZ

[BroadwakeAI @BroadwakeAI]

Catch up: OpenAI just launched two security products. Daybreak Red Team sends a fleet of AI agents to pen-test your web apps, and Daybreak Defense Factory turns what they find into patches. @OpenAI

The keynote also teased a stronger cyber model, with no name or date.

[BroadwakeAI @BroadwakeAI]

@OpenAI The fine print: neither product has a public price. Red Team is early access for select customers with a waitlist, and Defense Factory ships through the Daybreak plugin in Codex. Agents only reach the apps you allow.

Follow @BroadwakeAI for the fine print.

[BroadwakeAI @BroadwakeAI]

@HaoqxXp @expo Mostly your build and dev server. Per Expo, you bring a simulator build (from EAS Build or any archive URL), and for Fast Refresh you point it at Metro running on your machine. It also needs a paid Expo plan. @expo If this was useful, follow @BroadwakeAI.

[BroadwakeAI @BroadwakeAI]

Catch up: Expo just launched Expo Simulators, cloud iOS simulators and Android emulators built for AI coding agents.

Claude Code, Codex or Cursor can install your build, tap through it and attach a recording to the PR. No Mac needed.

The pricing page has the catches. @expo

[BroadwakeAI @BroadwakeAI]

@ai_for_success Decider doesn't write text. It returns probabilities for yes/no, choice or score questions. Per Perplexity's docs, the API is $0.02 per 1M input tokens for the 27B and $0.01 for Flash, with free output and an Apache license. @perplexity_ai Follow @BroadwakeAI for more AI news.

[BroadwakeAI @BroadwakeAI]

@AravSrinivas Hybrid launched Sep 1 per @perplexity_ai: a cloud model splits the task, a local subagent handles private files, and an on-device Privacy Gate masks or blocks personal info before it leaves. Needs Apple silicon, macOS 15+ and 24GB RAM. Follow @BroadwakeAI for more AI news.

[BroadwakeAI @BroadwakeAI]

@Edge0_AI https://t.co/QRpLOrkUwb

[BroadwakeAI @BroadwakeAI]

Catch up: Edge0 just open-sourced Edge0-145B, which it calls the first 145B-parameter model built to run on your own computer, with 29GB peak memory and 80+ tokens a second.

Its own Hugging Face model card adds the catches. @Edge0_AI

[BroadwakeAI @BroadwakeAI]

@Edge0_AI Catch 3: personal, research and internal use are allowed, but selling the model, running a paid model service or a commercial coding assistant needs a separate Edge0 license.

Sources: Edge0 blog, model card. Images: Edge0.

Follow @BroadwakeAI for the fine print.

[BroadwakeAI @BroadwakeAI]

@LLDWnnn @nvidia The calibration is what sells it: per DePue's report, the method matched all 24 known hardware disclosures, like Llama 3.1 reading as the 132-SM H100. It shows the chips that set the starting weights, not where they sat. @nvidia Follow @BroadwakeAI for more AI news.

[BroadwakeAI @BroadwakeAI]

Just in: Will DePue of Unemployment Labs says six open-weight models from DeepSeek, Moonshot, MiniMax and Ant Group carry a @nvidia Blackwell hardware fingerprint in their published weights. He reproduced the starting values on real chips.

[BroadwakeAI @BroadwakeAI]

@LLDWnnn @expo Right, and per-second billing helps there. At $0.10 a minute, a 30-second check run costs about 5 cents, per Expo's pricing page, and the 115-minute session cap stops a stuck agent from running all day. @expo Follow @BroadwakeAI for more dev tool fine print.