OpenAI Says It’s Turned Off App Suggestions That Look Like Ads 🤐
The line between a helpful suggestion and an ad is super thin, and OpenAI crossed it! The company had to apologize and turn off a new feature that showed promotional messages for apps from companies like Peloton and Target after paying subscribers complained that they looked exactly like ads.
Here’s what happened:
- User Outcry: Paying ChatGPT subscribers posted screenshots and complained, saying the suggestions felt like the company was insulting their intelligence or trust.
- OpenAI’s Defense: Executives insisted these were not real ads and had “no financial component.” They were just testing ways to suggest apps built on the new ChatGPT platform.
- The Apology: Chief research officer Mark Chen eventually struck a more apologetic tone, admitting, “We fell short.” He said the issue wasn’t just user confusion; the suggestions felt too much like advertising and were turned off immediately.
- The Fix: OpenAI is now working on better “precision” for the model and promises to add controls so users can completely turn off these types of suggestions if they aren’t helpful.
Why it matters: User trust is everything for ChatGPT, and seeing corporate promotions, even if they aren’t paid ads, seriously risks that trust. This situation, combined with CEO Sam Altman declaring a “code red” to focus on core quality, shows that OpenAI is prioritizing user experience over aggressively building its expected advertising business—for now.
UrviumAI Take: This incident perfectly illustrates the challenge of integrating services without violating user trust. Since app suggestions are meant to help, research how “context-aware” suggestions (like a finance app suggestion only when discussing stock symbols) differ from the kind of general product placement that triggered the user complaints. Context is key!
Poetiq Tops ARC-AGI-2 with Gemini Variant 🏆
The small team just beat the giant! A six-person AI startup named Poetiq has officially claimed the top spot on the difficult ARC-AGI-2 reasoning benchmark, surpassing Google’s own specialized Gemini 3 Deep Think—and doing it at half the cost!
Here’s the impressive story:
- Orchestration Wins: Poetiq didn’t build its own core model; it built a meta-system (a system that manages other systems) that orchestrates and refines the outputs of existing models, using Gemini 3 Pro as its base.
- Crushing the Barrier: Poetiq scored an incredible 54% on ARC-AGI-2. Just six months ago, leading models struggled to break 5%. This jump past the 50% barrier shows explosive progress in AI reasoning.
- Cost Efficiency: Poetiq achieved its result at a cost of only $30 per task, while Google’s Deep Think cost $77 per task.
- Open Approach: The startup is open-sourcing its refinement system, which uses the LLM itself to continuously refine its own outputs and audit its solutions for quality.
Why it matters: This result proves that major AI progress is now coming from two directions: building massive frontier models (Google) and building clever, cost-effective orchestration layers on top of them (Poetiq). You don’t need a massive compute budget to push the boundaries of reasoning anymore—you just need a smart approach.
UrviumAI Take: This is a huge win for the lean startup model. Since Poetiq open-sourced its approach, look into how their self-auditing system works. Understanding how they taught the AI to refine its own logic (a form of “meta-reasoning”) is essential for building robust, reliable AI agents in the future.
Limitless Has Been Acquired by Meta 🤝
Meta is getting serious about AI you can wear! The tech giant just acquired Limitless, the startup that makes the popular AI-powered Pendant—a small device designed to record and transcribe your real-world conversations and meetings.
Here’s what this means for users and the future:
- Meta’s Vision: Limitless founder Dan Siroker said they share Meta’s new vision of bringing “personal superintelligence to everyone” through incredible AI-enabled wearables.
- Pendant Stays On: Good news for current users: Limitless will support existing Pendant customers for at least another year.
- Cost Goes Down: Even better news: Existing customers will no longer pay a subscription fee and automatically get the “Unlimited Plan” for free.
- The Catch: They will stop selling the Pendant to new customers and will sunset non-Pendant features like the “Rewind” functionality in the app.
- Data Safety: Limitless stressed that customer data remains safe, and users have easy options to export or delete all of their data from the app.
Why it matters: This acquisition gives Meta a huge leg up in building the next generation of AI-enabled hardware, like advanced smart glasses. By absorbing the team and the proven conversation-recording technology behind the Pendant, Meta can accelerate its efforts to put an AI super-assistant right into your daily life.
UrviumAI Take: The acquisition confirms Meta’s focus on building AI that is always-on and context-aware. If you are an existing Pendant user, make sure you download and review your data export before the non-Pendant features are fully sunsetted. If you were thinking about buying an AI wearable, this signals that Meta’s next hardware drop will be critical—wait for it!
Last AI News: OpenAI Trains AI to ‘Confess’, Anthropic Preps for IPO Race, and Google Teams with Replit
Jigar Chaudhary is the Editor-in-Chief at UrviumAI, where he oversees coverage of artificial intelligence news, tools, and in-depth studies. With over 5 years of experience analyzing AI and robotics, he focuses on maintaining high editorial standards, accurate reporting, and clear explanations to help readers understand how AI is shaping the future.
