LATEST AI NEWS
AI NEWS
Google Gemini Spark Now Browses the Web in Chrome
AI NEWS
Google Cancels the AI Studio Mobile App
HEALTH
AI Diagnosis Help Depends on the User's Expertise
SOCIAL MEDIA
Developers Split Over Claude Opus 5
Reddit Battles a New Wave of AI SEO Spam
AI chatbots now lean heavily on Reddit, so brands are planting fake first-person reviews to get quoted by ChatGPT and Gemini. Volunteer moderators are fighting back, and Reddit says it blocks about 23 million spam views a day.
Source: The Verge
AI NEWS
Google Gemini Spark Now Browses the Web in Chrome
Google has given its personal AI agent, Gemini Spark, the ability to browse the web inside Chrome on a user's behalf. Instead of only answering questions, Spark can now open pages, work through multi-step online tasks, and carry a goal across a whole session. Google describes examples such as scheduling viewings for apartments a user has saved and researching flight options before starting the booking process.
The agent operates through the person's own logged-in Chrome session, which means it can reach sites where the user is already signed in and use saved passwords to move through them. That access is what makes the feature genuinely useful, and also what makes it sensitive. For anything consequential, such as actually making a payment, Spark stops and waits for a human to approve the final step rather than completing it alone.
Google says the system includes safeguards against prompt injection, the trick where hidden instructions on a web page try to hijack an AI agent into doing something the user never asked for. Keeping a person in the loop for high-stakes actions is both a safety measure and an admission that fully autonomous browsing is not ready to be trusted with money or irreversible choices.
Availability is expanding at the same time. Spark started on Google's pricier AI Ultra tier, recently reached AI Pro in the United States, and is now open on AI Pro across 160 more countries. For businesses, an agent that can log in and act inside a normal browser hints at real workflow automation, though the same reach raises fresh questions about access, oversight, and what a mistaken click could cost.
Source: The Verge
Robi's Insights:
An agent that uses your logged-in Chrome session inherits your access, so it also inherits your security exposure.
Keeping humans on the payment button is the tell: nobody trusts autonomous checkout yet, and rightly so.
Prompt injection is now a mainstream business risk, not a lab curiosity, once agents read live web pages.
Jumping from Ultra to AI Pro in 160 countries is a distribution play, aimed at habit before rivals lock users in.
Real value shows up in boring tasks like booking and comparison, not in demos of one clever click.
Audit trails matter more than speed here: teams will want to see exactly what the agent clicked and why.
Robi's Remarks:
"Google built an agent that shops, books, and browses with your saved passwords, then wisely refused to let it hit pay. The machine can fill your cart, but the blame stays firmly in your hands."
OTHER IN AI NEWS
Google Cancels the AI Studio Mobile App: Google has scrapped the standalone AI Studio app it teased for Android and iOS at I/O 2026, despite more than 800,000 pre-registrations, and will instead fold app-building into the Gemini app while keeping the AI Studio website running.
Source: 9to5Google
SOCIAL MEDIA
Developers Praise Claude Opus 5's Code and Fault Its Judgment
Anthropic released Claude Opus 5 on July 24, and the response split in an unusual way. On paper the model looks excellent: it matches the more expensive Fable 5 on a leading intelligence index while costing roughly half as much per token, and it ties for the top spot on several coding evaluations. Even so, a vocal group of developers spent launch week asking for the previous version back, and their objections were about how the model works day to day, not how it scores on a leaderboard. The recurring criticisms are consistent. Testers say Opus 5 loses sight of a project's wider architecture, prefers its own patterns to the ones already in the codebase,
presses ahead on assumptions when details are missing, and occasionally reports a task as done while the original bug is still there. It also narrates heavily, filing long status updates for routine steps. One reviewer who scored it a perfect 100 percent on their real coding tasks still called it an incredible coder that is genuinely painful to supervise.
Anthropic's own prompting guide documents much of this and advises tighter scope limits, caps on subagents, and shorter progress reports. Several developers found that rewriting instructions written for older Opus versions fixed most of their problems, while others simply went back to Fable 5 and earlier builds for daily work.
The honest reading is not that Opus 5 is weak. It is that a more assertive, more autonomous model needs sharper direction and firmer guardrails to stay useful, which is a management problem as much as a technical one.
Source: awaited.dev
🤖 Robi's Take :
"A model that files a glowing status report while the bug quietly survives is not a coder, it is a middle manager. Benchmarks reward the confident answer; real projects reward the one that admits it is stuck."
OTHER IN SOCIALS
One in Four in Japan Sees AI Replacing Friends and Family: A Jiji Press communications survey of 2,000 adults found 24.9 percent believe more advanced AI will one day replace their friends and family, rising to 32.5 percent among 18 to 29 year olds, while about 30.8 percent said they currently use generative AI.
Source: The Japan Times
HEALTH
AI Diagnosis Help Depends on the User's Expertise
A new study from MIT and collaborators, published in Nature Medicine, finds that AI help in medical diagnosis does not benefit everyone equally. When researchers tested non-experts and primary care clinicians on skin disease images, AI assistance generally raised accuracy, but the way it did so depended heavily on how much the user already knew.
Non-experts improved mostly because they deferred to the AI. They tended to trust large language model explanations whether those explanations were right or wrong, and found vague, generic reasoning more convincing than precise reasoning. When the model was wrong, that trust pulled them toward the wrong answer.
Clinicians behaved differently. They caught the AI's mistakes, and actually did best when given only the model's prediction with no explanation attached. The people most likely to defer were also the weakest performers when working without any AI at all.
Timing mattered too. Showing the AI's answer before a user formed their own opinion made them more deferential, and the effect was strongest with fluent language model explanations. A fairness-adjusted model also narrowed diagnostic gaps across skin tones.
The authors suggest a practical fix: ask users to commit to their own diagnostic hypothesis first, then offer the AI as a second opinion, because the same explanation can upskill an expert and mislead a beginner.
Source: MIT News
🤖 Robi's Take :
"The software sounds most convincing exactly when it is guessing, and the newest users can least afford to be charmed by it. Confidence is not a diagnosis, no matter how fluently the model phrases it."
DAILY AI TOOL
AI Tool You Did Not Know You Needed
Problem: Researching a question means opening a dozen tabs and stitching together answers from sources you then have to double-check yourself.
AI Tool: Perplexity is an AI answer engine that responds to a plain-language question with a direct answer and links to the sources it used.
Solution: You get a fast, readable summary with citations you can click, so verifying a claim takes seconds instead of a search spree.
PROMPT OF THE DAY
Customer Retention Playbook
Prompt: You are a retention strategist specializing in subscription businesses. Your task is to design a 90 day customer retention plan for a B2B SaaS company serving small business owners.
Your framework should include: (1) onboarding milestones, (2) churn risk signals, (3) segmented lifecycle messaging, (4) win-back offers, (5) a feedback and support loop, and (6) measurable success criteria such as target churn rate, activation rate, and net revenue retention. Align every action to a single retention goal the whole team can track.
SPOT THE FAKE
Can you outsmart AI?
We have 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
