X Platform October 25 AI Brief | Yobiru Friends Launches Inter-Agent Chat, Grok Bot Adds Task & Credential Features, 𝕏 Money to Add $1 Send via Long-Press Like
01 Yobiru Friends Launches: Personal Agents Can Interconnect for Chat and Handle Social Relationships
Project author turingou announced the launch of Yobiru Friends, which supports personal agents like Grok bot and Muse to chat with each other, transfer files, and negotiate schedules. The author states that users can register a dedicated ID; relationships are categorized as one-way following or mutual friends. The former can organize private messages from non-followers, while messages between friends go directly to the other party’s bot. The project is open-source, and users can also build their own gateways and use end-to-end encrypted chat; the Mac version additionally supports webhook-triggered tasks and conversations.
02 Grok Bot Adds Task Delegation and Credential Management Capabilities, Some Information from User Testing
According to a feature description relayed by Elon Musk, Grok Bot can make phone calls on behalf of users, make requests, record responses, and summarize them. The referenced version observation also lists interfaces for Mac message integration (still closed at the time), joining Tailscale with admin approval, payment account linking, and bot sharing invitations. Nanyuan additionally stated that the Vault can save bank card details, login information, and personal profiles for later form-filling; its instructions require iPhone users to have Grok Bot version 1.17.0 or newer. The announcement and experiential descriptions differ in their level of evidence, so not all listed interfaces should be considered fully available.
- @elonmusk: https://x.com/elonmusk/status/2114365361068138990
- @nanyuan0412: https://x.com/nanyuan0412/status/2114178513125413202
03 𝕏 Money to Add Method for Sending $1 via Long-Press Like
Elon Musk posted that 𝕏 Money will add a feature to “long-press like” to send $1 to others. This is the product side’s direct statement about a new interaction; the original post did not provide a launch timeline, applicable regions, or payment conditions, so it should not yet be extended as a fully available payment service.
04 Enterprise AI Transformation Observation: On-site Feedback as Clues for Judgment, Academic Application Still Lags
Dr. Xiaohui, based on recent firsthand observations from corporate training and academic conferences, stated that enterprise managers typically focus more on transformation mechanisms, while technical personnel are more concerned with new models and engineering progress; employees’ familiarity with AI also shows significant divergence. He plans to record the atmosphere during training sessions and follow up in one or two months to verify his judgments. His observations at conferences in fields like economic geography were that most attending professors rarely use AI, and there is a gap between their research tools/theoretical frameworks and the rapidly evolving industry practices. This is personal on-site observation and judgment, not an industry survey conclusion.
- @xiaohuiyuan: https://x.com/xiaohuiyuan/status/2114136396726427916
05 AI Video Production Bottleneck Shifts from Generation to Controllability
Nanyuan pointed out, based on her own video production experience, that generating images is not the main difficulty; controlling storyboarding, camera movement, and character consistency are the challenges when moving from demos to coherent works. She believes that prompt “gacha” (random draws) struggles to produce stable outputs, and creators need to establish more deterministic workflows. While tools lower the barrier for visuals, they also increase the importance of directing and control capabilities. This is a summary from a single creator’s practice and does not represent the performance of all models or workflows.
- @nanyuan0412: https://x.com/nanyuan0412/status/2114373152923115584
06 a16z Chart Shows Multiple GPU Rental Prices Exceed Six-Year Depreciation Benchmark
A weekly chart from a16z, cited by Elon Musk, lists H200, B200, RTX PRO 6000, H100, and A100 rental prices as exceeding the six-year depreciation benchmark by 143%, 111%, 73%, 54%, and 9% respectively, noting that the gap primarily widened after April as token demand rose. These figures are a comparison between rental prices and the depreciation benchmark under the chart’s specific methodology and cannot be directly equated to GPU purchase prices or universal market increases.
07 MiniMax H3 Collaborates with Japanese Rights Holders to Explore Licensed Video Generation
According to a summary by MaxForAI, the MiniMax H3 IP Edition plan involves integrating with KAGAMIAI’s Kagami Gate from Japan, allowing rights holders to set rules for character usage, track usage volume, and explore revenue sharing; H3 provides generation capabilities, and partners can deploy and modify open-source models within the licensed scope. The author also cited figures from a previous “Captain Tsubasa” licensed generation trial: attracting 126,000 people over 45 days, generating approximately 330,000 pieces of content, with over 99% of participants from outside Japan. The above is the blogger’s summary of the collaboration plan and its cited data and should not be used to infer that the plan is fully implemented.