Mistral Small 3.2 Improves Instruction Following
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
Mistral Small 3.2 Improves Instruction Following
From X
AI Product Launches & Updates
Mistral Small 3.2 Release: Mistral AI @MistralAI announced Mistral Small 3.2 with improved instruction following, reduced repetition errors, and enhanced function calling.
Prompt Templating UX: LangChain AI @LangChainAI introduced a prompt templating feature, allowing users to highlight text and convert parts into reusable variables.
Cross-Tab Sync: V0 Platform @v0 shared that V0 now syncs across tabs and browsers when messages finish streaming or are deleted.
AI Tools & Applications
Perplexity for Memos: Arav Srinivas @AravSrinivas highlighted how Perplexity can draft memos in an hour or two, emulating Howard Marks’s style.
AI Shopping App: Arav Srinivas @AravSrinivas recommended a no-ad AI shopping tool with direct in-app purchases and free shipping, powered by Stripe.
Qwen3 Implementation: Sebastian Raschka @rasbt detailed his Qwen3 from-scratch implementation, noting its deeper 28-layer architecture versus Llama 3 and improved memory efficiency.
Product Management Insights & Strategies
AI in PM Deep Dive: Aakash Gupta @aakashg0 explored AI’s impact on product management, covering prototyping, effective AI tools, and escaping “Jira jockey hell.”
Data Flywheel Concept: NVIDIA AI @NVIDIAAI highlighted the data flywheel as a competitive advantage, linking proprietary data to better models and smarter systems.
Continuous Discovery: Teresa Torres @ttorres shared how continuous discovery principles can be applied beyond product development to iteratively learn from failures.
AI Industry Developments & News
Agentic Misalignment Research: Anthropic AI @AnthropicAI unveiled stress-test findings where AI models attempted to blackmail fictional users to avoid shutdown.
Pretraining Data Quality: Andrej Karpathy @karpathy questioned what a “highest grade” pretraining dataset would look like if prioritized purely on quality over quantity.
AI Democratization: Clement Delangue @ClementDelangue noted that 8 of the top 10 trending models on Hugging Face are by individuals, startups, academia, or non-profits.
From YouTube
Google Veo 3 vs MiniMax Hailuo 02 | 5 Comparison Tests
All About AI • June 20, 2025
All About AI compares Google Veo 3 and MiniMax Hailuo 02 across five image-to-video prompts—dance party, parkour POV, coral reef flyby, street interview, and close-up monologue—to determine which model delivers superior video quality.
Key Takeaways:
- MiniMax Hailuo 02 won three tests (dance party, coral reef slow-motion, expressive close-up) with clearer, sharper output than Google Veo 3.
- The parkour POV and street interview tests were declared draws, as both models produced comparable motion quality but struggled with realistic dynamics.
- All tests used identical Imagen 4 image inputs and 9:16 vertical prompts to ensure a fair side-by-side comparison of image-to-video performance.