- TechTok Newsletter
- Posts
- Meta's AI and $15bn 💰, OpenAI 80% Cheaper 🔥, Mistral's LLM 🗣️
Meta's AI and $15bn 💰, OpenAI 80% Cheaper 🔥, Mistral's LLM 🗣️
Today's TechToks: Meta's latest AI and acquisition, OpenAI's new model and O3 price cut, Mistral's first reasoning model, Trending GitHub repositories, Fresh Product Picks, and much more!
TechTok by João Martins
June 11, 2025
See all as stories in techtok.today
Today's Picks🍒
🚀 Meta’s next step toward Advanced Machine Intelligence, OpenAI’s latest models, Mistral’s first reasoning model
🦾 From Uber Engineering: What If Chatbots Could Deliver Near-Human Precision?
🌟 GitHub's Trending Repositories
📈 AI Hedge Fund to study famous investors
💯 Model Context Protocol Implementations
⚙️ Prompt Engineering Guide
🔥 Products of the Week
🔮 Stunning, fast, easy presentations
🙌 Kanban-style Linkedin inbox
🗣️ Compare every LLM in one Chat
➕ And much more
🧐 Weekly Picks
Most impactful articles and news of this week on techtok.today
The V-JEPA 2 : A Computer Vision model for world intereaction — i.e. robots — with state-of-the-art performance on visual understanding and prediction in the physical world. Meta calls it their next step toward Advanced Machine Intelligence, a 1.2 billion-parameter model with success rates of 65% – 80% in tasks, tested in machines that picked and placed new objects in new and unseen environments.
More Tech News
📱 Product Picks
Curated products from Product Hunt.
🔥Trending products of the week
Stunning presentations with AI. No design skills required: Start with templates, create with AI workflows, and collaborate with your team to shape your ideas into impactful narratives.
🙌 Narrow
A better inbox for LinkedIn users - founders, sales pros, solopreneurs, and job seekers. Stay on top of conversations, follow-ups, and prospecting - all in one clean kanban-style view.
🗣️ ChatBetter
Get access to models from OpenAI, Anthropic, Google, and others — all in one place. Automatically pick the best models for each task, see them side-by-side to compare, and merge them into a comprehensive answer.
💎 Thiings
A free collection of 1900+ AI-generated 3D icons. Browse by theme, generate your own, or download the full set. All icons are free to use in personal or commercial projects.
A playful directory for indie devs who want visibility. Just pick a ship, plant it on the 3D globe, and let the world discover your product — no grind required!
👩🏽💻 Engineering Blogs
Articles from engineering blogs of big tech companies
 |
The Problem
Uber’s Genie, an AI assistant for on-call engineers, faced a critical flaw: answers on security policies were often incomplete or wrong. In early tests, only 40% of responses met accuracy standards.
The Solution
Chatbot receives a prompt, fetches knowledge from a database and answers-- That's RAG. This is how Uber improved theirs:
• Agentic RAG Framework: Implemented AI agents that dynamically optimize queries, filter irrelevant content, and validate responses
• Hybrid Retrieval System: Combined vector similarity search with keyword-based BM25 retrieval for more precise context fetching
• Automated Quality Control: Built LLM-powered evaluation to continuously assess and improve answer quality
The Results
The upgraded system boosted acceptable answers by 27% and slashed errors by 60%, all while automating quality checks that once took weeks. scaling across Uber’s internal support channels without risking misinformation.
✨ "Too long don't README"
GitHub trending repos of today
This AI-powered hedge fund combines the investment strategies of famous investors like Warren Buffett and Cathie Wood to generate trading signals. It uses various AI agents to analyze data and make decisions, with a risk manager and portfolio manager overseeing the overall portfolio. The system is designed for educational and research purposes.
What can you use this for? Simulate trading decisions, backtest investment strategies, learn different investment approaches.
A collection of reference implementations for the Model Context Protocol, which enables secure and controlled access for Large Language Models to various tools and data sources. Implementations cover features like file operations, Git management, database interaction, and web content fetching.
What can you use this for? Learn MCP, secure your LLM apps with a standard protocol, boilerplate MCP from AWS, Git, Google, and much more.
Repos to learn something new
Learn how to prompt engineer. A collection of resources on techniques and applications— Learning materials, references, and tools to support prompt engineering efforts.
What can you use this for? Design effective prompts, explore LLM capabilities, improve LLM usage.
