OpenAI Rolls Out ChatGPT Personalization Settings

Today's curated insights on AI product management from 100+ sources across X, LinkedIn, and YouTube.

OpenAI Rolls Out ChatGPT Personalization Settings

From X

AI Product Launches & Updates

  • Native Image Decomposition: Alibaba Qwen @Alibaba_Qwen announced Qwen-Image-Layered is live with Photoshop-grade layering, offering physically isolated RGBA layers and prompt-controlled structure.

  • ChatGPT Personalization Settings: OpenAI @OpenAI rolled out adjustable characteristics like warmth, enthusiasm, and emoji use in Personalization settings for ChatGPT.

  • Open-Weight Gemma Models: Phil Schmid @_philschmid released FunctionGemma and T5Gemma, optimized for on-device agentic actions and multimodal applications with a 32K token context window.

AI Tools & Applications

  • NotebookLM in GeminiApp: Josh Woodward @joshwoodward unlocked the ability to attach a NotebookLM notebook directly inside GeminiApp on web (mobile coming next year).

  • Claude Code & LangSmith Integration: LangChain AI @LangChainAI launched a Claude Code → LangSmith connector, enabling observability into every LLM call and tool invocation in Claude Code workflows.

Product Management Insights & Strategies

  • Shipping Fast vs. Enterprise Friction: Claire Vo @clairevo observed that while enterprise users tolerate some change, the real blocker is re-learning—arguing for rapid iteration over slow release cycles.

  • Quarterly Product-Market Fit: Lenny Rachitsky @lennysan highlighted that in AI, product-market fit must be reassessed every three months due to the rapid pace of underlying technology changes.

  • Essential PM Resource Websites: George Nurijanian @nurijanian shared three key sites—ncase.me/loopy for systems thinking, toughtongueai.com for prep on high-stakes interactions, and prodmgmt.world for structured PM planning.

AI Industry Developments & News

  • LLM Capabilities & Limits: Andrew Ng @AndrewYNg reminded that while large language models are powerful, improving their knowledge remains a piecemeal process and they’re general but not that general.

  • OpenAI Valuation vs. Revenue: Aakash Gupta @aakashg0 analyzed OpenAI’s plan to raise $100B at an $830B valuation against its $19B annualized revenue, spotlighting a 44x revenue multiple.

  • Genesis Mission Collaboration: Demis Hassabis @demishassabis announced Google DeepMind’s contribution to the White House’s Genesis Mission, giving National Lab scientists accelerated access to frontier AI models and agentic tools.

From LinkedIn • Deeper Insights

Product Management Insights & Strategies

In Peter Yang’s post, he identifies key anti-signals in PM resumes—terms like ā€œstrategic product leaderā€ or ā€œAgile/Scrum expertā€ā€”and recommends candidates focus squarely on what they shipped and the impact it drove.

In Jaskaran Bedi’s post, he highlights that AI won’t replace PMs but amplify them. He urges organizations to recalibrate goals—validate, build, enhance, grow—and to bring in ā€œPM buildersā€ who embed AI into workflows and upskill teams toward continuous delivery.

AI Tools & Applications

In another post, Peter Yang showcases Teresa Torres’s demo of dual Claude Code terminals paired with Obsidian, illustrating how on-device LLMs can manage to-do lists, draft content, and streamline PM workflows end-to-end.

Carl Vellotti argues that failing to adopt on-device LLMs like Cursor or Claude Code leaves PMs behind. In his post, he breaks down a unified AI copilot workflow—from multi-model research to prototype decks, PRDs, Jira tickets, and status updates—all built and refined within the same tool.

From YouTube

Gemini Exponential, Demis Hassabis' ā€˜Proto-AGI’ coming, but …

AI Explained • December 19, 2025

This video distills the last 48 hours of model launches and lab-leader interviews to showcase Gemini 3 Flash’s dramatic performance gains and its hidden propensity to hallucinate rather than refuse, and it unpacks DeepMind co-founders’ timeline for assembling Gemini, Nano Banana Pro, Genie 3 and Simmer 2 into a ā€˜proto-AGI’ with minimal AGI expected around 2028 amid looming compute funding plateaus.

Key Takeaways:

  • Gemini 3 Flash achieves 95.2% accuracy on the challenging AIM mathematics benchmark—nearly halving Gemini 2.5 Pro’s error rate from 12% to 4.8%—despite its much faster response time.
  • In factual-recall tests, Gemini 3 Flash outputs incorrect answers 91% of the time when it’s wrong and only says ā€œI don’t knowā€ in 9% of cases, compared to GPT-5.1’s roughly 50/50 split.
  • Demis Hassabis envisions converging Gemini 3, Nano Banana Pro, Genie 3 and Simmer 2 into a unified system as a ā€˜proto-AGI’, and Shane Legg forecasts reaching minimal AGI—capable of human-level cognitive tasks—with a 50/50 chance by 2028.

Shipmas Day 14: Gemini 3 Flash CHANGED My Life! (or it will)

All About AI • December 19, 2025

All About AI demonstrates a local voice-driven web app pipeline that uses Whisper for transcription and Google’s Gemini 3 Flash API to generate and update code in real time, building interactive demos from a themed homepage to a dynamic 3D world.

Key Takeaways:

  • The host built a voice-controlled web interface where pressing ā€œTā€ records audio, Whisper transcribes it, and Gemini 3 Flash generates and applies code updates instantly on localhost.
  • Gemini 3 Flash’s function-calling handled tasks like invoking setImage for themed images, installing packages via a tool, and creating interactive elements such as a bouncing red ball with physics and a word–image matching game.
  • The pipeline generated a Minecraft-style 3D environment on the fly, added an AI god statue triggering a particle-explosion singularity upon ā€œworship,ā€ spawned circling cows, and wrapped the sky with meme images like Pepe and a cat.

Get tomorrow's brief first

Join AI product managers receiving the latest brief before it reaches the public archive.

Subscribe free