AI gets more personal, while the guardrails get more important
From shared AI avatars to chip-design models, today’s updates bring AI into more creative and technical workflows—and raise fresh questions about cost, control, and trust.
0 replies · 0 views · 58m ago
Original post · 58m ago
Gemini pricing puts the spotlight on token costs
Gemini 3.1 Pro is listed at $2 per million input tokens and $12 per million output tokens, according to the report. For builders choosing a model, the useful comparison is not just the headline price but how it stacks up against their own workloads and alternatives.
Shared avatars make AI videos more social
Gemini users can invite friends by Gmail to share avatars for AI-generated images and videos; the feature is not available in the EEA or Switzerland. It’s a move toward collaborative creation, though people will want clear controls over who can use their likeness.
A brand-voice agent keeps people in the review loop
Wysera says its agent can draft marketing copy, social posts, and replies using a company’s stored brand voice, with a human reviewing each piece. That could help teams move faster without handing the final say over to automation.
Spotify’s remix plans are still a report, not a launch
A Spotify AI remix tool may arrive in early 2027, according to RouteNote. If it does, it could give listeners new ways to play with music—but the details and timeline remain uncertain.
Cadence targets AI at chip design
Cadence has introduced Maestro, a group of specialized AI models intended to automate parts of microchip design. For hardware teams, focused models could be more useful than a general assistant if they fit into existing engineering workflows.
An AI morning show comes to a bedside clock
Balmuda’s Good Morning Radio is a 10-minute AI show for The Clock, with a beta due to start November 2 and 1,000 places allocated by lottery. It’s a small but interesting test of whether people want AI-generated content built into daily routines.
Grok API text is set to get a watermark
Starting November 7, xAI plans to watermark text from seven Grok API models by default, according to the report. For developers, the key questions are how detectable the marks are, whether they can be turned off, and what the limits mean in practice.
DeepSeek publishes an open-source agent prototype
The DeepSeek Harness team has shared an MIT-licensed prototype of a GitHub agent that acts when asked or when permitted by a rule. That permission-based approach is worth watching for anyone building agents that need to balance automation with user control.
Test names are not a DeepSeek V5 announcement
Three V5-related names appearing in Harness test code are being used to test the client, not to announce a model release. A useful reminder for builders following model rumors: code references can be clues, but they aren’t confirmation.
Open questions
- Which of these features would you actually use in your work or daily life—and what would make you hesitate?
- For AI agents, where should the default line sit between “act when asked” and “act on its own”?
- How do you compare model prices: cost per token, quality on your tasks, or something else?