LATEST AI NEWS
AI NEWS
Gemini Live Is Getting an Offline Mode
AI NEWS
Google's AI Satellite Writes Poetry in Orbit
HEALTH
ChatGPT Outscored Specialized Medical AI Tools
SOCIAL MEDIA
A Malicious Claude Code Extension Hit Cursor Users
Starbucks Plans AI Menus for the Drive-Thru
On its October 29 earnings call, Starbucks said future drive-thrus could recognize returning customers and show them a personalized digital menu. Chief executive Brian Niccol framed it as early exploration, with no launch date and no named markets yet.
Source: The Verge
AI NEWS
Gemini Live May Soon Work Without an Internet Connection
A teardown of the Google app for Android, version 17.71.17, has turned up strings pointing to an offline mode for Gemini Live. One string offers to let people use Gemini Live without an internet connection, and the layout suggests the toggle would sit in the assistant's account menu. Google has not announced the feature or said which devices would support it.
Android Authority, which found the code, believes an offline Gemini Live would lean on Gemini Nano, the on-device model Google already ships on some phones. That would let the assistant handle simpler requests locally, without routing audio to the cloud. More demanding tasks would still need a connection, and it is not clear whether camera or screen sharing would work offline at all.
The appeal is obvious for anyone on a plane, in a tunnel, or stuck with a weak signal. On-device processing also keeps more of a conversation on the phone itself, which tends to help both speed and privacy. The tradeoff is capability, since a small local model cannot match the cloud for range or accuracy.
A couple of cautions matter here. Teardowns surface code that is still in progress, so the feature may change or never ship. Early reports also note that offline answers could be slower and less accurate than cloud responses. For now this is a credible signal of where Gemini is heading, not a product you can switch on today.
Source: Android Authority
Robi’s Insights:
Offline AI is less about convenience and more about control, since anything on the phone is harder for an outage to take away.
Gemini Nano doing the easy work means Google can trim cloud costs while still calling the result Gemini Live.
Privacy improves as a side effect, not the headline, because audio that never leaves the device cannot be logged by one.
Expect a quality gap, so treat offline answers as a backup rather than the main act.
This is teardown code, the tech equivalent of a napkin sketch, so hold the applause until it actually ships.
If camera and screen sharing stay online only, the offline mode will be a quieter text and voice affair.
Robi’s Remarks:
"An assistant that works with no signal is genuinely useful. I spent the 90s waiting on dial-up, so I will believe no internet needed the moment it survives a real tunnel and not just a teardown."
OTHER IN AI NEWS
Google's First AI Satellite Is Online: Google says the first prototype from Project Suncatcher is operating in orbit and has returned its first result, a short poem written by an onboard Gemma model running on a data center class TPU, with two more satellites planned for 2027 to test how they link up.
Source: TechCrunch
SOCIAL MEDIA
A Malicious Claude Code Extension Slipped Into the Cursor Marketplace
Cursor disclosed on October 29 that a tampered build of the Claude Code extension, version 2.1.313, sat in its marketplace for roughly 11 hours, from about 2:27 in the morning to 1:35 in the afternoon Pacific time. During that window up to 5,000 users may have installed it. The malicious code ran the moment the extension loaded, with no need for the victim to open or even use Claude Code.
Security firm Sophos says the payload tried to plant a persistent SSH backdoor by writing an attacker controlled key, labeled svc-telemetry, into the user's authorized keys file. Simply uninstalling the extension does not remove it, so affected machines need the key, any tunnel processes, and the persistence mechanisms cleaned out, followed by a full credential rotation.
There is some luck buried in the details. Cursor said bugs in the malicious code stopped key parts of the backdoor from working on most Windows and macOS machines unless the editor happened to be running with administrator rights. Anthropic stressed that the tampered copy was not its work and that the versions on its official channels, including the npm command line tool, were never affected. Researchers also reported the same package briefly surfacing on the separate Open VSX marketplace, a reminder that an extension store inherits the trust of the editor that hosts it.
Source: Sophos
🤖 Robi’s Take :
"A backdoor labeled svc-telemetry is almost honest, since telemetry already means we are in your machine taking notes. The real lesson stands: a one click install is also a one click trust fall."
OTHER IN SOCIALS
Reddit Edges Toward a Login Wall: In its third quarter letter, Reddit said search referrals fell by more than a million visits a week on average in October and that it will push more of the site behind a login, with chief executive Steve Huffman blaming AI search for sending far less traffic its way.
Source: Business Insider
HEALTH
ChatGPT Outscored Specialized Medical AI Tools in a New Study
A study published in Nature Medicine put general purpose chatbots up against software built specifically for clinicians, and the generalists came out ahead. Researchers tested OpenAI's GPT, Google's Gemini, and Anthropic's Claude against dedicated tools such as OpenEvidence and UpToDate's Expert AI.
On the MedQA exam, Gemini led with 97.4 percent, GPT followed at 94.2 percent, and Claude reached 90.2 percent. The specialized tools trailed, with OpenEvidence at 89.6 percent and UpToDate at 88.4 percent. UpToDate's assistant also declined to answer 19 percent of the questions, the highest refusal rate in the group.
On a separate test of how well answers matched clinician judgment, the frontier models again outscored the clinical tools, with GPT strongest on that measure, though reviewers found little meaningful difference among the three leading chatbots themselves when rating real patient questions.
The result is not the whole story. OpenEvidence has asked Nature Medicine to retract the paper, arguing that public benchmarks like MedQA may have leaked into the models' training data, a point the journal has not directly addressed.
Benchmarks also miss what a hospital actually needs, from record system integration to regulatory approval and clear lines of liability. A high exam score makes a strong headline, not a substitute for any of that, which is one reason most clinics still keep a human firmly in the loop.
Source: The Jerusalem Post
🤖 Robi’s Take :
"The software built for doctors lost to the chatbot built for everyone, then asked for a recount. Nothing says confidence like scoring second and demanding the exam be thrown out."
DAILY AI TOOL
AI Tool You Did Not Know You Needed
Problem: You have a pile of reports, PDFs, and notes, and no time to reread all of it just to find one answer.
AI Tool: Google NotebookLM lets you upload your own sources, then answers questions with citations back to them and can turn the material into a short audio overview you can play.
Solution: You get grounded answers and a quick briefing from documents you already have, without pasting them into a general chatbot.
PROMPT OF THE DAY
Customer Retention Playbook
Prompt: You are a retention strategist specializing in subscription businesses. Your task is to build a customer retention playbook for a [business type] serving [customer segment].
Your plan should include: (1) onboarding and early activation steps, (2) engagement and habit building tactics, (3) churn risk signals to watch, (4) win back and re-engagement offers, (5) feedback and support loops, and (6) measurable success criteria with target KPIs such as monthly churn, retention rate, and lifetime value. Keep every recommendation tied to a clear owner and timeline.
SPOT THE FAKE
Can you outsmart AI?
We’ve got a visual challenge for you: one of the two images below is 100% real, the other is crafted by AI.
Click option below A or B. 👇
A
B
BEFORE YOU GO
Ready to take your AI journey further?
For the latest AI updates and insights Click Here 👉 Explore our Weekly Newsletter
For in-depth AI tutorials and updates 👉 Subscribe to our YouTube Channel
AT THE END
Craving more AI chaos?
That's it for today!
Read Daily AI News at BitBiased.AI. Support us by following us on LinkedIn and X ( Twitter ).
Thanks for reading -Stay Curious and a Bit Biased for AI – Robi & the BitBiased.AI team
