Perplexity Max Releases Agentic Comet Browser
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
Perplexity Max Releases Agentic Comet Browser
From X
AI Product Launches & Updates
Comet browser release: Aravind Srinivas @AravSrinivas released Comet, the first agentic browser designed as a thought partner, now available to Perplexity Max subscribers in an invite-only rollout.
MedGemma multimodal model: Jeff Dean @JeffDean unveiled MedGemma, an open-weights multimodal model for interpreting EHR data, medical text, and imaging across modalities.
Reachy Mini orders open: Clement Delangue @ClementDelangue announced opening orders for Reachy Mini, an open-source desktop robot for AI builders starting at $299, integrated with LeRobotHF and Hugging Face.
AI Tools & Applications
Local model reasoning support: LangChain Team @LangChainAI launched langchain-ollama with reasoning capabilities for local LLMs.
Snowflake RAG pipeline tutorial: Llama Index Team @llama_index introduced a guide to build a complete RAG pipeline using LlamaParse with Snowflake Cortex for enterprise document processing.
Bolt & Expo App Store workflow: Bolt Team @boltdotnew shared how @imalittledev used Bolt and Expo to streamline App Store submission.
Product Management Insights & Strategies
Product operating model change: Teresa Torres @ttorres shared that implementing a product operating model requires a complete organizational change and offered strategies to handle stakeholder resistance.
Balancing growth & UX: Teresa Torres @ttorres recommended the article “Growth at What Cost?” on the tension between revenue goals and user experience.
AI Industry Developments & News
Google Search's AI scale: Logan Kilpatrick @OfficialLoganK noted that Google Search stands as the largest AI product in the world.
Franco-British AI collaboration: Demis Hassabis @demishassabis recapped a discussion with President Macron on international AI collaboration to accelerate scientific discovery.
AI browser war emerges: There's An AI For That @theresanaiforit highlighted the AI browser war, featuring products like Perplexity Comet and upcoming competitors.
From YouTube
I'm REVEALING ALL the Vibe Marketing Secrets (NO Gatekeeping)
Greg Isenberg • July 09, 2025
In this episode, Greg Isenberg defines vibe marketing as the convergence of AI agents and vibe coding tools and walks through setting up MCPs in Code, plus detailed workflows for automating short-form content, one-click outreach, voice AI sales agents, and more.
Key Takeaways:
- Vibe marketing relies on AI agents like Sonnet 4 and Opus plus MCPs in Code to automate goal-oriented tasks—e.g., emailing a weekly business health dashboard by connecting to Google Drive and Airtable.
- One workflow scraps Reddit into Airtable, filters for relevant posts, uses V3 or Runway ML to generate 60-sec videos, stitches them with Creatomate, and auto-posts to social to scale meme-style accounts and drive traffic.
- Using Gum Loop’s browser extension, you can build a one-click CRM that scrapes X profiles, analyzes and enriches data, crafts personalized outreach, logs contacts in Airtable, and sends weekly summaries for sales, recruiting, or podcast bookings.
Learn to post-train LLMs in this free course
Deeplearning.ai • July 09, 2025
Ban Huatu introduces Deeplearning.ai’s free post-training course on LLMs, covering Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and online reinforcement learning methods (PPO & GRPO) to transform pre-trained models into instruction-following assistants.
Key Takeaways:
- Supervised Fine-Tuning (SFT) involves training a pre-trained base model on input-output pairs to signal shifts in behavior, such as following instructions or adopting a specific tone, demonstrated by fine-tuning a small base model into an instruction-following model.
- Direct Preference Optimization (DPO) shapes model behavior using a contrastive loss on paired responses—one preferred and one rejected—and is applied in the course to reshape a small instruct model.
- Online reinforcement learning techniques, including PPO and the more efficient GRPO, use rewards from human feedback or automated signals (e.g., math checking, unit tests), with GRPO applied in this course to improve a small Quinn model’s math skills.
This Google Veo 3 Update is a GAME CHANGER: Here's Why
All About AI • July 09, 2025
This video details Google’s new VO3 update, showcasing the frames-to-video feature with speech support on custom images, V3 fast cost savings, and improved audio handling, and demonstrates how to create consistent multi-scene AI videos using Context Pro.
Key Takeaways:
- Google VO3’s frames-to-video now supports speech in the first frame, allowing any uploaded image to become a talking head clip in V3 fast.
- The V3 fast option uses just 20 credits per generation (versus 100 credits for quality mode) and includes enhanced audio coverage, fewer unwanted subtitles, and bug fixes.
- Combining V3 fast with the Flux Context Pro image-to-image model enables consistent character continuity across scenes, simplifying workflows for multi-scene vlogs.