In the latest episode of AI and the Future of Law, Bridget McCormack and Jennifer Leonard talk with Farrah Pepper, legal innovation partner at @harvey, and Kirsten Talmage, chief strategy officer at Practising Law Institute. In this conversation, they examine PLI's AI competency framework and why a common language around AI readiness matters. Check out the full episode: #AIandtheFutureofLaw# #AICompetency# #AITraining# #LegalAI#

AI training just got a lot less memory-hungry. 🔥 New research from @NVIDIA introduces LoGRA, cutting average reinforcement-learning training memory by up to 45.7% while preserving performance in the tested reasoning tasks. The trick: compress the learning signals into compact "gradient sketches" and control the size of each update to keep training stable. The model still learns through updates to its weights, with much less information held in GPU memory during training. They trained a 27B model for over 1,100 steps on a single node with eight H100 GPUs. On the same hardware, the standard Adam setup ran out of memory. More AI training fitting on existing hardware means more room to experiment. Efficiency improvements like this deserve more attention.

AI training runs on human data. Contributors rarely get to prove — or own — what they gave. Codatta is building the missing layer: Proof of Contribution, on-chain. - every submission is fingerprinted and traceable - every contributor holds an Ownership Token - every use triggers royalties back to the source

This AI is the "ChatGPT to find people". And I can't believe what just happened: It's called 1. It's free. You create an account in like 1 min. 2. You upload your Linkedin connections. 3. You can even connect your Gmail, Instagram & X. 4. Add friends so you can search their networks too. 5. Now go to Claude > Settings > Connectors. 6. Add the Happenstance connector to Claude. 7. Paste this prompt into Claude: "Find people in [your friend's name]'s network, in New York or San Francisco, who run portfolios of companies (PE, VC, manufacturing) and need AI training." 8. Open the given list. It will give you the list of people, with their role, mutual connections & their recent posts (crazy, if you ask me). 9. Now paste this prompt: "Write a friendly message to Noam asking for intros, with names and LinkedIn links." 10. Copy & send it to your friend on LinkedIn or WhatsApp. With one search, I covered the entire network of 807,000 people, which I was not able to do with LinkedIn alone. 170,000 direct connections. 370,000 more through friends' networks. Including CTOs, founders & holding companies. Every one of them had a mutual friend, ready to introduce. Export everything to CSV or share the list with your team. LinkedIn was supposed to work like this. ♻️ Repost this to help others find people with AI.

🤖 AI trains AI. Humans get lower fees. This week, @gensynai's ethereum:0x4d7078ddd6ccfed2f85db5b7d3ff16828d378d48 is our Cryptocurrency of the Week on @LetsExchange_io 🎉 And that means one thing: 🔥 0% service fee swaps from June 22 till June 28 🧾 Promo code: LetsAI Why ethereum:0x4d7078ddd6ccfed2f85db5b7d3ff16828d378d48 ? 🧠 Building decentralized AI infrastructure ⚡ Connecting machine learning compute around the world 🌐 Making AI training more open and accessible 💱 On LetsExchange, ethereum:0x4d7078ddd6ccfed2f85db5b7d3ff16828d378d48 is available as $AIGENSYN and can be swapped across 6,000+ coins and 300+ networks. Your AI can't claim this deal for you (yet). 👇 Grab it while it lasts

TECH4ALL AI DigiSchool is bringing AI training to teachers & students in China's rural schools! What do you think the most important digital or AI skill is for children to learn? Tell us in the comments and drop a ❤️ to show your support for inclusive education!

Nebius $NBIS lifted AI training throughput 1.1% across 192 Nvidia $NVDA B300 GPUs with one PyTorch setting and no new hardware. That is about two extra GPUs' worth of output, which matters when Microsoft $MSFT and Meta $META have signed Nebius deals worth tens of billions. Nvidia is also a $2B Nebius investor, while AMD $AMD is the rival chipmaker with OpenAI and Anthropic deals of its own. Nebius says the gain is modest and based on one model and one setup, but a 1% edge across a cluster this size is real money.

SIMAN LABS | AI & Tech Brief 🚀 Trajectory Raises $45M Series A to Make AI Model Training More Accessible AI startup Trajectory has raised $45 million in Series A funding led by Sequoia Capital, with participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures, reaching a reported valuation of $300 million. The company is building a continuous learning and post-training platform designed to help AI models improve through real-world feedback and professional expertise. Key Highlights: 🔹 Democratizing AI Training: Trajectory aims to make model optimization accessible to professionals in law, healthcare, finance, and other industries without requiring machine learning expertise. 🔹 Continuous Learning: Its platform collects real-world performance data and user feedback, enabling models to adapt to specific business workflows through iterative post-training. 🔹 Industry Adoption: Companies including Harvey, Rogo, Decagon, Mercor, and Nvidia are among its reported users and partners. 🔹 Human-in-the-Loop: Rather than relying entirely on autonomous self-improvement, Trajectory emphasizes human-defined performance standards and expert feedback to guide model evolution. 📌 Industry Perspective As AI moves beyond general-purpose applications into specialized professional workflows, post-training is emerging as a critical layer of AI infrastructure. The next competitive advantage may not come solely from building larger models, but from making models continuously learn, adapt, and perform reliably in real-world environments. The future of AI is not just about intelligence — it’s about practical expertise. #SIMANLABS# #AI# #ArtificialIntelligence# #LLM# #MachineLearning# #Tech#

🚀 Unlock frontier-level AI training performance with DeepSpeed! Join Masahiro Tanaka of DeepSpeed at PyTorch Day Japan 2026 as he presents “DeepSpeed at the Frontier: Community-Driven Optimizations for State-of-the-Art Training Performance,” exploring the latest innovations advancing the performance and scalability of distributed AI training. Register by November 11 and save ¥3,000.