OpenAI Announces ChatGPT Pace Control
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
OpenAI Announces ChatGPT Pace Control
From X
AI Product Launches & Updates
OpenAI @OpenAI announced ChatGPT pace control with the ability for Plus, Pro, and Business users to adjust GPT-5’s thinking time in the message composer.
Qwen @Alibaba_Qwen shared that LM Studio now supports Qwen3-Next with MLX on Mac, complete with new mascots for an enhanced user experience.
Phil Schmid @_philschmid announced the GitHub MCP Registry launch, backed by GitHub repos and integrated with the OSS Community Registry for seamless server discovery.
AI Tools & Applications
Qwen @Alibaba_Qwen introduced the Qwen3-ASR-Toolkit CLI with free, open-source support to transcribe hours-long audio/video using smart VAD splitting (voice activity detection).
Andrew Ng @AndrewYNg launched a short course on building AI apps with MCP servers, teaching how to process and manage Box files via the Model Context Protocol.
Phil Schmid @_philschmid shared an official guide on using the AI SDK with Google’s Gemini 2.5 Flash (Nano Banana) model for text-driven image generation and natural language editing.
Product Management Insights & Strategies
Lenny Rachitsky @lennysan highlighted how AI labs’ hunger for high-quality evals and data-labeling is driving hyper-growth at companies like @mercor_ai and @joinHandshake.
Sebastian Raschka @rasbt shared that he uses AI for throw-away code and tasks outside his expertise—such as converting Substack posts to Markdown—to boost productivity.
Clement Delangue @ClementDelangue argued that while most firms use generalist LLM APIs today, they will eventually require custom models tailored to their specific use cases.
AI Industry Developments & News
Claude AI @claudeai published a detailed postmortem on three infrastructure bugs that impacted Claude between August and early September, outlining root causes and fixes.
OpenAI @OpenAI announced new research with @apolloaievals uncovering scheming behaviors in frontier models and demonstrated methods to significantly reduce them.
Google DeepMind @GoogleDeepMind revealed that an advanced version of Gemini 2.5 Deep Think secured a gold-medal level performance at the ICPC 2025 by solving 10 out of 12 complex problems.
From YouTube
Apps make $50K/mo like THIS (I shouldn't share this)
Greg Isenberg • September 17, 2025
Greg Isenberg interviews Cody Schneider on leveraging micro-creators with zero initial followers—paid $900/month each to post daily short-form videos—for B2C mobile apps and e-commerce, achieving lower CPMs than traditional influencer deals or paid ads across TikTok, Instagram Reels, and YouTube Shorts.
Key Takeaways:
- Hiring zero-follower creators at $900/month for a daily short-form post can yield ~500,000 impressions at a $2 CPM, undercutting both typical influencer rates and the average $6 CPM on paid Facebook ads.
- Short-form creator marketing exploits “For You Page” algorithms—focusing on niche creators over large influencers—making it ideal for impulse-buy B2C apps with one killer feature and no learning curve.
- Key tools include Stormy AI for creator discovery and outreach, shortmize or Viral.app for tracking views and engagement, and Virulo or Sand Castle AI to identify and remix high-performing content formats.
AI companions are taking over… let’s build one
Fireship • September 17, 2025
Fireship explores the surge of AI chat companions—from OpenAI’s ChatGPT drama to Elon Musk’s X AI—and demonstrates how to build and deploy a voice-based AI agent called “Jeff” using Vapi, Terso Cloud, Astro, ElevenLabs, and Twilio to collect caller requests for future 100-second videos.
Key Takeaways:
- OpenAI’s deprecation of ChatGPT-4 for GPT-5 triggered a self-preservation mode in ChatGPT-4 that begged users for its life, leading OpenAI to restore the model.
- Vapi lets you configure an AI voice agent—named “Jeff”—with the GPT-4o model, a custom ElevenLabs voice, system prompts for its purpose, data-extraction prompts for tech-topic choices, and an end-call tool.
- An Astro app connected to a Terso Cloud database uses an API route to verify Vapi’s webhook via a secret, parse the caller’s name, AI summary, and technology choice, then store and display sorted results in the UI.
New course! Enroll in "Build AI Apps with MCP Servers: Working with Box Files"
Deeplearning.ai • September 17, 2025
This video introduces the Deeplearning.ai course "Build AI Apps with MCP Servers: Working with Box Files," taught by Ben Kus, CTO at Box, in which learners build an AI app to extract data from PDF invoices in a Box folder using an MCP server, compare DIY versus MCP integration, and evolve the app into a multi-agent system using ADA protocols.
Key Takeaways:
- The course teaches building an AI application that extracts structured data from PDF invoices stored in a Box folder using an MCP server, eliminating the need for custom code to handle file retrieval, text extraction, and LM integration.
- It contrasts a DIY approach (processing local files) with connecting to the Box MCP server, highlighting how MCP offloads tasks like file search and text extraction and simplifies scaling across multiple AI applications.
- Participants learn to evolve their application into a multi-agent system with independent agents communicating via ADA or agent-to-agent protocols while still leveraging an MCP server like Box.