Oct 9 · AI Daily

Meta chief AI officer Wang Tao stated that users who want the Muse agent to help them earn $10,000 still need to put in effort themselves, and Meta is continually improving Muse's memory capabilities to handle more complex tasks. A month after its launch, Muse remains at the top of the Apple App Store free apps chart, and Meta is exploring sustainable business models for it such as subscriptions and transaction commissions, without spending much time considering ads for now.

An article introducing how to build automation using Claude Managed Agents, with the example of deploying an agent scheduled to read Slack and GitHub, equipped with credentials and memory, and publishing updates. Setup can be completed with a single Claude Code command, and API credits included in the Max and Team plans can be used for this purpose. Original link: https://claude.dev/blog/building-effective-agent-automations/

A Databricks blog post introduces how to use Lakebase Postgres database branching capabilities to support Agentic SDLC. Lakebase can branch a database of any size in under a second, using copy-on-write to share parent branch data and supporting scale-to-zero.

Databricks released Funke, a Python and PySpark library plus deployable pipeline that parses HL7v2 messages directly into native Spark types while preserving full hierarchical structure. It is the successor to Smolder, an open-source Scala data source released in 2021, supporting all HL7 message types and versions, ingested through Declarative Pipelines and stored in Unity Catalog.

Listening to another episode of the Arena Alignment Index breakdown by @ml_angelopoulos today on @tbpn

We are thrilled to announce that all superintelligent organizations have now jointly agreed on the ultimate solution for AI/SI safety:

Moving all tests to Delta Airlines flights, where getting online is strictly impossible!

According to the Financial Times, OpenAI disclosed to investors that its annualized revenue approached $50 billion by the end of September, significantly lower than the previously estimated $70 billion by outsiders, leaving a gap of about $20 billion. The difference stems from varying statistical criteria, as Anthropic includes sales revenue from partners like AWS and Google Cloud, while OpenAI excludes it. Following the report, US tech stocks fell, with the Nasdaq 100 index closing down 1.4%, Nvidia down 2.9%, and Oracle down 5.5%.

Excited to support @WhiteHouse, @ENERGY, and Director @mkratsios47 this week with a $150 million investment in the Genesis Mission. Looking forward to collaborating on the future of science!

When it comes to physics, who has the best open-source video world model? 💫

It's me again, MiniMax H3. Almost on par with closed-source sota. 🍾

According to The Verge, Anthropic has revised its Claude usage policy, taking effect on November 12, 2026, adding for the first time a clause prohibiting "sustained and unnecessary abuse or cruelty" toward the model, with the primary enforcement method being conversation termination. This clause does not apply to common complaints, questioning, dark creative themes, and model testing and research. The policy also expands weapon bans, prohibits use for surveillance and law enforcement target decisions, and requires autonomous physical hardware to be supervised by qualified operators.

Anthropic published an interview with Bradley Axen, head of Block AI capabilities, detailing how Block used Claude Fable 5 to orchestrate large-scale code migrations: Fable handles high-level design such as data models and API specifications, and then coordinates dozens of smaller Opus or Sonnet models to execute file modifications and testing, with a single migration potentially merging thousands of pull requests.

The open-source Atomic Agent Desktop app is now available on macOS, Windows, and Linux, running open models like Qwen and Gemma locally for free without requiring an account, and connecting to cloud models only when chosen by the user.

Anthropic launched OSS Scanner, allowing open-source projects to opt in and receive free regular security scanning reports generated by its strongest models (including Claude Mythos). The reports are entirely model-generated with no human review, which may find vulnerabilities faster and more frequently, but may also contain errors or invalid reports. The report also noted that AI tools have helped discover major open-source security vulnerabilities in recent months, while some open-source projects (including Linus Torvalds and Google) are struggling to cope with the influx of AI-generated vulnerability reports.

Artificial Analysis evaluated the newly released low-cost video model Grok Imagine Video 1.5 Lite from SpaceXAI, which ranked 17th on both the AA-Video-T2V v2.0 and AA-Video-T2V-Silent v2.0 leaderboards, priced at about one-third of Veo 3.1 while ranking ahead of it.

Databricks introduced the LTAP architecture to Lakebase Postgres, offloading batch data loading from the primary compute node to the Spark distributed engine, with internal benchmarks showing that loading 1 TB of data now takes less than 5 minutes.

Hugging Face engineer yuvraj sharma used HuggingChat's ML Intern mode to build 7 models in a few days, including a 0.8B prompt rewriter that runs on CPU with 99.7% effective output, and a Qwen3.5-2B fine-tuned model that improved citrus disease recognition accuracy from 14.9% to 52.8%, with a total computing cost of about $103.

The AMIE research conducted by Google in collaboration with BIDMC was published in The Lancet, marking the first prospective study of patient-facing conversational diagnosis in a real outpatient setting. AMIE is a research system that patients can converse with prior to their visit. In tests with real emergency patients, clinicians reported that summary feedback helped them prepare for visits 75% of the time, and it influenced the diagnostic approach in over half of the cases; AMIE's differential diagnosis matched the doctors' final diagnosis 90% of the time. This is also the first time Google has published a paper in the main issue of The Lancet.

Ethereum researcher Justin Drake suggested the crypto industry should prepare a "bunker mode", arguing that AI progress could crack the Elliptic Curve Digital Signature Algorithm (ECDSA) used to protect wallets within the next few months. Vitalik Buterin expressed support for this view.

Google announced that its medical research system AMIE is the first patient-facing conversational diagnostic tool to undergo prospective study in a real clinical environment, with results published today in The Lancet. The study showed that patients conversing with AMIE before face-to-face visits boosted their confidence and helped organize their thoughts; doctors saved time by reviewing patient conversation information, allowing them to focus more on collaborative care and joint decision-making.

This research, conducted in collaboration with @BIDMC_Medicine, demonstrates the potential of this technology in strengthening the doctor-patient relationship.

Learn more ↓ https://goo.gle/3VCGhMl