ChatGPT Image Generation on WhatsApp
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
ChatGPT Image Generation on WhatsApp
From X
AI Product Launches & Updates
ChatGPT image generation on WhatsApp: OpenAI @OpenAI announced that ChatGPT image generation is now available in WhatsApp via 1-800-ChatGPT, enabling users worldwide to create images directly in their chats.
Qwen3 models launch: Alibaba Qwen @Alibaba_Qwen launched Qwen3 models in MLX format with 4-bit, 6-bit, 8-bit, and BF16 quantization, optimized for the MLX framework.
Model sizeābased selection feature: Clement Delangue @ClementDelangue released a new feature on Hugging Face that helps teams find the best model based on size to balance performance and cost.
AI Tools & Applications
Ephemeral tokens for Gemini Live API: Philipp Schmid @philschmid highlighted support for ephemeral tokens in Google DeepMindās Gemini Live API, enabling secure, short-lived WebSocket connections for real-time agent interactions.
Perplexity Tasks integration: Aravind Srinivas @AravSrinivas announced integration of Perplexity Tasks with Finance Watchlist, allowing custom natural language alerts and workflows for investment portfolios.
AI Travel Agents multi-agent demo: LlamaIndex š¦ @llama_index showcased Microsoftās AI Travel Agents demo using the Model Context Protocol, LlamaIndex.TS, and Azure AI Foundry to coordinate six specialized agents for complex travel planning.
Product Management Insights & Strategies
Multi-agent system strategy: Harrison Chase @hwchase17 discussed when to build multi-agent systems, linking Anthropicās blog on their research setup and Cognitionās post on avoiding multi-agent architectures, offering strategic guidance for PMs.
Meeting dynamics advice: Lenny Rachitsky @lennysan shared that every additional participant in a meeting adds cost and reduces candidness, urging PMs to include only essential stakeholders for effective problem-solving.
AI Industry Developments & News
3D asset generation model: Rowan Cheung @rowancheung shared an AI news roundup that highlighted Tencentās cinematic-quality 3D asset generator and other key updates, reaching over 511,000 views.
LLM prompt injection risks: Andrej Karpathy @karpathy warned about prompt injection attacks in LLMs, comparing the threat landscape to early computing malware and calling for stronger defenses.
Sabotage evaluation suite: Anthropic AI @AnthropicAI unveiled a new evaluation suite that tests models on hidden āsabotageā tasks alongside benign objectives, helping developers monitor and mitigate emerging risks.
From YouTube
Perplexity's new AI agents are INSANE
Greg Isenberg ā¢ June 16, 2025
Greg Isenberg tests Perplexity Labsā AI agents by creating an 8-minute mini-film storyboard and script for ideabrowser.com, building a Fortune 500 and YC founder sales lead dashboard, and generating a 90-day social media growth strategy based on Dan Koeās framework to showcase the toolās real-time workflows.
Key Takeaways:
- Perplexity Labs generated six detailed storyboards with sketches, character profiles (e.g., Maya Chen, a 24-year-old McDonaldās employee), lighting directives, and a full screenplay for an idealbrowser.com ad.
- The AI agent compiled a dashboard listing 20 Fortune 500 executives and Y Combinator foundersācomplete with emails, phone numbers, HQ addresses, company intros, filters by priority, and personalized outreach templates.
- It produced a 90-day content calendar to reach 100,000 followers by emulating Dan Koeās hook-to-CTA framework, outlining niche analysis, pillar distribution (40% educational, 25% inspirational, etc.), daily hooks, CTAs, and budget estimates.
How AI Video Is Taking Over The Internet
All About AI ā¢ June 16, 2025
This video explores how AI tools like V3, 11 Labs and HeyGen are driving down costs and fueling viral contentāfrom a $2,000 AIāgenerated NBA Finals ad to rapidly growing ASMR channels on X and automated avatars made from a single photo.
Key Takeaways:
- The $2,000 fully AI-generated NBA Finals ad used V3 to produce 15 usable clips from 300ā400 iterations, slashing production costs by 95% compared to traditional ads.
- An AI-generated ASMR channel on X gained 82,000 followers in three days and 13 million views by leveraging V3ās native audio synthesis capabilities.
- HeyGenās Avatar feature can animate a single uploaded photo with an 11 Labs-generated voiceover, creating a cohesive lip-synced video from just one image and audio file.
How to use Cursor for interactive prototypes, sound design, and data visualization
How I AI Podcast ā¢ June 16, 2025
Independent design educator Elizabeth Lin demonstrates how to use Cursorās conversational AI to generate and iterate on interactive web prototypesāfrom prompting it to suggest and apply design styles, to scaffolding a playable MacOS-style piano with sound, to refining a finance dashboard by removing drop shadows and emulating top product designs.
Key Takeaways:
- Designers can prompt Cursor with "What design aesthetics and movements are you comfortable implementing? List the styles and describe them" to receive descriptive keywords (e.g., neon color schemes, glitch effects) and then reprompt it with styles like Brutalist or Y2K to automatically redesign a blank homepage.
- By defining a simple Cursor ruleā"if I ask you to create a new prototype, copy this folder and then add it to my homepage"āLin generates fully functional interactive prototypes, such as a digital piano in the style of old MacOS that responds to clicks and produces different waveforms from a single prompt.
- Lin iteratively transforms an initial ugly finance dashboard by sequentially instructing Cursor to remove all drop shadows, simplify its color palette inspired by Edward Tuy, and "improve the layout to make it look like something a top designer at Apple would approve of," resulting in a polished, data-driven UI.