AI Weekly Update - October 11

Tesla's Robotaxi drives up, AMD and Nvidia trade blows in chip wars

what to know for now

🤖 Tesla unveils Cybercab robotaxi. Tesla introduced Cybercab, a robotaxi featuring neither steering wheel nor pedals, priced under $30,000, arriving by 2026. Robovan, designed to carry 20 people or cargo, was also revealed; pricing and release date remain undisclosed. Read more

🏥 Microsoft brings AI to healthcare. Microsoft announced AI tools to help doctors and nurses by automating administrative tasks. These tools include medical imaging models, a healthcare agent service, and a documentation solution for nurses. Read more

🧑‍💼 Zoom introduces AI avatars. Users can create custom AI avatars from recorded videos to send messages and attend meetings. The feature includes advanced authentication, watermarking, and is available via a $12/month add-on. Read more

🧠 Google AI wins Nobel Prizes. Demis Hassabis and John Jumper received the Chemistry Nobel for decoding protein structures, while Geoffrey Hinton was awarded Physics for machine learning advancements. The awards highlight Google's leading role in AI research and ignite debate on recognition and regulatory pressures. Read more

🧪 AI Research of the Week

MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering
From OpenAI

Jake’s Take: The paper introduces MLE-bench, a benchmark designed to evaluate the machine learning engineering capabilities of AI agents. It uses 75 curated Kaggle competitions, assessing skills like model training, dataset preparation, and experiment execution. The benchmark measures AI performance against human baselines, with the best-performing agents reaching medal-level in 16.9% of competitions. Resource scaling and pre-training contamination impacts are also analyzed, contributing insights into agent autonomy and its limitations in real-world tasks.

This paper will help push the AI industry to re-evaluate the gap between human and machine capabilities in practical ML engineering, hopefully fostering competition that will drive more robust AI agents developed responsibly.

what to know for later

🚀 AMD launches AI chip rival. AMD introduced the Instinct MI325X, an AI chip to compete with Nvidia's Blackwell GPUs. The MI325X begins production in 2024, aiming to challenge Nvidia's data center GPU market share. Read more

📈 Blackwell AI processors sold out. Nvidia’s next-generation AI chips are fully booked for the next twelve months, with production progressing on schedule. Analysts maintain buy ratings, forecasting strong 2025 performance. Read more

🎯 AI enhances Naval gunfire. DARPA's Heimdall project integrates AI and machine learning into the Mk 45 deck gun's Mk 160 computer system to automate target selection and improve accuracy. The program includes development and integration phases with in-lab and at-sea testing. Read more

🔒 Industry unready for agent security. Enterprise AI agents require secure identities and access management across multiple systems. Experts indicate existing authentication frameworks are inadequate, recommending API security models for agent identity protection. Read more