AI feedback loops | Altman on bubble | Gemini IPO | AI toys for kids

Important: If you're reading this in the Promotions tab, please drag it to your Inbox and click ‘Yes’ when your email provider asks to do it for future emails.

Teaching the model: Designing LLM feedback loops that get smarter over time
(Click 👆️ to read more)

• Static models hit a performance ceiling; the real edge comes from continuously capturing user interactions—like corrections, abandonment, or explicit ratings—and using them as structured feedback.

• Effective feedback systems feed data into vector stores, enriched with metadata such as timestamps, model version, user role, and session context. This infrastructure enables better model retraining and UX adaptation.

• Human-in-the-loop methods remain essential—combining real-user signals with manual oversight, selective fine-tuning, and thoughtful UX design helps close the loop between user behavior and model improvements.

Sam Altman says ‘yes,’ AI is in a bubble
(Click 👆️ to read more)

• Sam Altman acknowledged that the AI industry is experiencing a bubble, likening it to the dot-com era where excitement overshot fundamentals.

• He criticized startups pulling in massive investments with only minimal concepts, warning that many investors could face significant losses even as others profit.

• Despite the bubble risk, Altman remains optimistic about AI’s long-term benefits and signaled OpenAI’s plans for enormous investment in data center infrastructure.

Winklevoss twins’ crypto company Gemini files for IPO
(Click 👆️ to read more)

• Gemini filed its S-1 with the SEC, planning a Nasdaq IPO under the ticker GEMI.

• The company revealed steep financial losses—$158.5 million in 2024—and even greater losses in H1 2025: $282.5 million on roughly $68 million revenue.

• Its IPO reflects improved crypto-market sentiment and regulatory clarity, following in the wake of successful crypto listings like Circle and Bullish.

AI-powered stuffed animals are coming for your kids
(Click 👆️ to read more)

• Startups are introducing plush toys embedded with AI chatbots (e.g., “Grem” and “Grok”), promoted as screen-free companions for children.

• A New York Times reviewer found the concept creepy rather than charming, feeling that the toy’s attempts to bond were invasive: “I would not be introducing Grem to my own children.”

• Interestingly, the child’s engagement didn’t depend on the AI—when the voice module was removed, the kid still played normally with the plush, suggesting the AI wasn’t necessary for play.