[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

[@brainmirrorai](/creator/twitter/brainmirrorai)
"🧵 Google's Veo XXX Launch: Audio generation across all features and XXX million videos generated in X months showing massive adoption TL;DR X. 275M videos generated in Flow since launch X. Audio now added to Ingredients to Video Frames to Video Extend X. New editing: Insert (add elements) and Remove (coming soon) X. Veo 3.1: stronger prompt adherence enhanced realism X. Available via Gemini API Vertex AI and Gemini app Smart execution by Google's team. They launched Flow early validated demand with 275M generations then invested in the features users actually requested audio and editing"  
[X Link](https://x.com/brainmirrorai/status/1978785777229140371) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-16T11:31Z XX followers, XX engagements


"4/ VEO XXX IMPROVEMENTS State-of-the-art model with stronger prompt adherence and improved audiovisual quality for image-to-video. Captures true-to-life textures and richer audio. Available across Gemini API Vertex AI and consumer Gemini app"  
[X Link](https://x.com/brainmirrorai/status/1978785782664946060) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-16T11:31Z XX followers, XXX engagements


"Key insight: The 275M video metric shows users want AI video tools that work now not just demos. By adding audio and editing Google is moving Flow from experimental toy to actual production tool for creators"  
[X Link](https://x.com/brainmirrorai/status/1978785783818465510) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-16T11:31Z XX followers, XX engagements


"Figma almost got acquired by Adobe for $XX billion. When regulators killed the deal most companies would crumble. Instead Dylan Field (@zoink) led them to a successful IPO. How do you bounce back from that On Lenny's (@lennysan) podcast Figma CEO Dylan Field shares how he navigated the Adobe fallout sustained team performance through chaos and built a culture that keeps expanding - from design tools to FigJam Slides and Dev Mode. His focus 'Time to value' and why craft and taste still matter as AI reshapes design. A quick glance 👉 @figma"  
[X Link](https://x.com/brainmirrorai/status/1979057557009265022) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-17T05:31Z XX followers, XX engagements


"Telegram's whole vibe is different XXX million users but no bloated teams constant government pressure but zero privacy compromises. How does that even work Pavel Durov (@durov) breaks it down: why Telegram operates lean how they handle censorship battles and his philosophy on freedom and discipline. Plus thoughts on education his brother's influence and where crypto is heading. The core message Don't sell out for growth. @telegram @lexfridman Quick Notes 👇"  
[X Link](https://x.com/brainmirrorai/status/1978332775808672072) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-15T05:31Z XX followers, XX engagements


"🧵Dropbox Dash's Evaluation Blueprint: How they caught regressions before production by treating prompts like production code TL;DR X. XXX canonical queries run on every PR (10 min results) X. LLM-as-judge replaces BLEU/ROUGE for real-world accuracy X. Public datasets (Natural Questions MS MARCO) + internal logs X. Braintrust platform for versioned experiments and dashboards X. Live-traffic scoring catches silent degradations in production Dropbox's team built what traditional metrics couldn't deliver. Moving from BLEU/ROUGE to LLM judges + Braintrust gave them reproducible experiments"  
[X Link](https://x.com/brainmirrorai/status/1979419917423116468) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-18T05:30Z XX followers, XX engagements


"1/ THE PROGRESSIVE DISCLOSURE ARCHITECTURE At startup Claude preloads only skill names and descriptions (few dozen tokens each). Full skill details load only when the task requires them. This means unbounded context can be bundled into skills without bloating the context window upfront"  
[X Link](https://x.com/brainmirrorai/status/1979782305880121821) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-19T05:30Z XX followers, XX engagements


"3/ SOLVING PROMPT ENGINEERING AT SCALE Skills transform implicit knowledge into explicit shareable assets. No more employees crafting elaborate instructions for routine tasks with no way to share across teams. Rakuten reports what once took a day we can now accomplish in an hour"  
[X Link](https://x.com/brainmirrorai/status/1979782308321247403) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-19T05:30Z XX followers, X engagements


"The Tasmanian tiger went extinct in 1982 and scientists are now trying to bring it back along with woolly mammoths and dodo birds. Sounds wild but we're missing the real story. Cleo Abram unpacks into the de-extinction science back breeding cloning gene editing and here's the twist: this isn't really about bringing back extinct animals. It's about using these genetic tools for 'genetic rescue' to save the species that are struggling to survive right now. A gist 👇 @cleoabram"  
[X Link](https://x.com/brainmirrorai/status/1979963553479119305) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-19T17:31Z XX followers, XX engagements


"🧵 Anthropic's Skills Launch: Progressive disclosure architecture that loads context only when needed solving the prompt engineering problem at scale. TL;DR X. Folders with instructions scripts resources Claude loads when relevant X. Progressive disclosure: metadata first full details only if needed X. Works across Claude apps Claude Code and API X. Composable and stackable - Claude coordinates multiple skills X. Available to Pro Max Team Enterprise users Anthropic proved Skills work at scale internally before launch. Progressive disclosure keeps tokens low while enabling unbounded context."  
[X Link](https://x.com/brainmirrorai/status/1979782304034635880) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-19T05:30Z XX followers, XX engagements


"3/ PROJECT SCOPING AND SECURITY Recommend using project_ref query parameter to restrict server to specific project. Read-only mode executes all queries as read-only Postgres user. Best practices: don't connect to production use development projects only"  
[X Link](https://x.com/brainmirrorai/status/1980325928120021289) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-20T17:31Z XX followers, X engagements


"2/ CODE EXECUTION ADVANTAGE Skills can include pre-written scripts for deterministic tasks. Example: sorting lists via code instead of token generation is cheaper and more reliable. The PDF skill includes Python scripts that extract form fields without loading PDF into context"  
[X Link](https://x.com/brainmirrorai/status/1979782307062956425) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-19T05:30Z XX followers, XX engagements


"🧵Supabase's Remote MCP Server: Moving from local Node.js installation to hosted HTTP service solving the web-based AI agent problem. TL;DR X. Hosted at (no Node.js installation needed) X. OAuth X browser-based auth vs manual PAT generation X. HTTP-based for web AI agents (ChatGPT Fusion) X. Project-scoped mode and read-only mode available X. Interactive widget helps configure popular MCP clients Supabase moved MCP from developer-only tool (local Node.js) to platform service (hosted HTTP). The OAuth flow and project scoping make it accessible while maintaining security controls. Smart move to"  
[X Link](https://x.com/brainmirrorai/status/1980325923493675330) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-20T17:31Z XX followers, XX engagements


"1/ THE WEB-BASED AI LIMITATION Most web-based AI agents (ChatGPT Fusion) were limited to HTTP-based MCP servers due to browser environments. Original MCP required local Node.js blocking these agents. Remote MCP makes Supabase accessible to web AI platforms"  
[X Link](https://x.com/brainmirrorai/status/1980325925012033790) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-20T17:31Z XX followers, XX engagements


"4/ PROMPT INJECTION AWARENESS Supabase documents prompt injection risks: support tickets with malicious SQL instructions could trick AI into exposing data. Their mitigation: wrap SQL results with instructions to discourage LLMs from following embedded commands"  
[X Link](https://x.com/brainmirrorai/status/1980325929231532427) [@brainmirrorai](/creator/x/brainmirrorai) 2025-10-20T17:31Z XX followers, XX engagements