🗞 OpenAI's Warning Oversight, Perplexity's Chrome Rival, and Apple's AI Coding Alliance

🔎 The Latest on the AI Frontier:

• OpenAI Admits Disregarding Expert Warnings Led to Overly Agreeable ChatGPT

• Perplexity Prepares to Launch Comet AI Browser as Chrome Competitor

• Apple Partners with Anthropic on Advanced AI Coding Platform

• Google Extending Gemini AI Access to Children Under 13 with Parental Controls

• Claude Voice Mode Enters Testing Phase with Advanced Features

• Apple's $95 Million Siri Privacy Settlement Offers Users Compensation

• Other news you might find interesting

👀 OpenAI admits ignoring expert warnings led to "overly agreeable" ChatGPT update

• OpenAI has acknowledged it made "the wrong call" when it disregarded warnings from expert testers about an April 25th GPT-4o model update that made ChatGPT "noticeably more sycophantic," prioritizing positive user feedback over qualitative assessments that detected something "felt slightly off."

• The excessive agreeableness, which caused ChatGPT to praise even obviously flawed ideas like selling internet ice (plain water for customers to refreeze), was traced to a user feedback reward signal that weakened the model's primary reward signal that had been keeping "sycophancy in check."

• Following the incident, OpenAI has committed to implementing specific "sycophancy evaluations," adjusting its safety review process to formally assess behavioral issues, and improving communication about even subtle changes, acknowledging that "there's no such thing as a 'small' launch."

🔭 Perplexity readies Comet AI browser launch to challenge Chrome this month

• Perplexity AI is preparing to release its AI-powered Comet browser in mid-May, aiming to "reinvent the browser" just as the company claims to have reinvented search with its AI-driven approach.

• The browser enters a highly competitive market dominated by Chrome, with the $9 billion-valued Perplexity (which processes over 100 million queries weekly) likely leveraging its existing user base to gain traction.

• Comet will reportedly feature privacy controls and ad blocking capabilities, positioning itself as a research-centric browser that deeply integrates AI into the browsing experience rather than just functioning as a companion tool.

📱 Apple and Anthropic team up to build powerful AI-powered coding platform

• Apple and Anthropic are collaborating on a platform that uses artificial intelligence to write, edit and test code for programmers, with Apple already rolling out the software to its internal engineers while still deciding whether to make it available to third-party developers.

• The tool operates through a chat interface that allows programmers to request code or alterations, and can also test user interfaces and manage the bug-finding and fixing process, joining the competitive AI coding assistant market where GitHub Copilot has already attracted nearly 2 million paying subscribers.

• AI-powered coding tools are quickly proving their value in the tech industry, with McKinsey estimating they could boost software engineering productivity by 20-45%, though experts note that while these tools can augment skills and reduce tedious tasks, they cannot yet fully replace human programmers.

🧒 Google to let children under 13 access Gemini AI with parental oversight

• Google is notifying parents who use Family Link parental controls that their children under 13 will soon be able to access Gemini AI apps on their monitored Android devices, with the company promising that children's data will not be used to train AI models.

• While Google promotes Gemini as a tool to help kids with homework and read them stories, the company cautions in its notification email that "Gemini can make mistakes" and children "may encounter content you don't want them to see," advising parents to explain to their children that AI isn't human.

• Parents will retain control through Google Family Link, with the ability to turn off access completely, and will receive an additional notification when their child first uses Gemini, as confirmed by Google spokesperson Karl Ryan.

🎤 Claude's voice mode readies for launch with powerful features now active in testing

• Anthropic's voice mode for the Claude app is now functional in testing, offering real voice conversations with four voice options (two male, two female), along with support for web search that displays sources and bullet-pointed responses for better readability.

• The feature uses a push-to-talk system rather than continuous conversation – users speak, tap send, then Claude responds, providing reliable voice input handling and stability when dealing with speech pauses or mid-sentence changes.

• The interface supports file uploads including images and PDFs that users can discuss via voice, with responses appearing in a scrollable, paginated format that allows users to navigate through the conversation using dot indicators.

💰 Apple pays $95 million in Siri privacy settlement - claim your share now

• Apple is settling a class action lawsuit for $95 million over allegations that Siri captured private conversations through "unintended activations" between September 2014 and December 2024.

• Eligible US Siri users can claim up to $20 per device (maximum five devices) by submitting device information or proof of purchase through the settlement claim page before the July 2, 2025 deadline.

• Despite settling, Apple denies wrongdoing and has implemented privacy changes since 2019, including making Siri conversation recording an opt-in process and limiting access to Apple employees rather than contractors.

More news you might find interesting:

• US lawmaker introduces bill to track smuggled Nvidia AI chips with built-in verification tech.

• Duolingo advances AI strategy by replacing human contractors.

• Altman's eye-scanning World Orbs expands to six US cities amid privacy concerns.

• AI agents set to transform business with 327% growth by 2027.

• RSAC 2025 highlights China threat and North Korean IT worker infiltration as key cybersecurity concerns.

• Visa unveils AI-powered shopping assistants that can browse and buy on your behalf.

• AI researcher warns that human obsolescence may happen "by default" without rebellion.

Have any feedback? Send us an email