Hugging Face just launched the Open CoT Leaderboard to measure how much chain-of-thought prompting actually boosts LLM accuracy, rather than tracking absolute benchmark scores. Early results from 30 models reveal small 7B models often gain more from step-by-step reasoning than giants, though CoT can occasionally backfire and degrade accuracy. It gives developers a real look at prompt effectiveness beyond raw capability.
Microsoft AI CEO Mustafa Suleyman just delivered a TED Talk offering an eye-opening mental model for artificial intelligence, arguing that we are essentially creating a "new digital species." He pointed out that even leading AI creators struggle to define where the technology is heading, urging the world to look closely at this pivotal moment.
Researchers just open-sourced JAT, a multi-purpose Transformer agent built to reproduce and improve upon Gato. Running on a single network, it tackles 157 tasks across Atari games, navigation, and robotic control, matching 65.8% of expert performance while supporting basic text and vision tasks. Why it matters: It opens up generalist embodied AI research with fully public models and datasets.
DARPA just revealed that an AI-controlled fighter jet successfully went head-to-head against a human pilot in an in-air dogfight. Flying the experimental X-62A against a human-piloted F-16, the two jets reached speeds of 1,200 miles per hour and got within 2,000 feet, with human safety pilots never needing to take back control.
Microsoft just rolled out industrial AI solutions designed to bridge factory-floor data with generative tools, including a factory operations Copilot template on Azure AI and data integration in Microsoft Fabric. Front-line workers can use natural language to troubleshoot malfunctioning machinery and query operations in real time. Core features are currently in private preview.
Meta chief AI scientist Yann LeCun just poured cold water on AGI hype, arguing that large language models will never reach human-level intelligence. He claims LLMs lack fundamental traits like reasoning, planning, and physical world awareness, suggesting the industry must pivot toward sensor- and video-trained "world models" instead.
OpenAI just dropped a majorly improved GPT-4 Turbo model in its API and began rolling it out to ChatGPT. GPT-4 Turbo with Vision is now generally available, finally letting developers use JSON mode and function calling directly on visual inputs. Why it matters: Multimodal workflows can now return reliable, structured data straight from images.
Stability AI just launched Stable LM 2 12B, a multilingual model trained on 2 trillion tokens across seven European languages. Available in both base and instruction-tuned flavors, it is built to handle function calling and RAG workflows on standard hardware. The compact 1.6B model also got an upgrade with better conversation skills and tool use.
Google DeepMind just introduced SIMA, a generalist AI agent capable of navigating 3D virtual environments and video games using only screen pixels and natural language instructions. Trained across titles like No Man’s Sky and Teardown, SIMA executes over 600 basic skills via standard keyboard and mouse controls. It marks a shift from chasing high scores to building versatile virtual assistants.
Deepgram just launched Aura, a real-time text-to-speech API designed to help conversational AI agents speak without awkward pauses. It renders realistic voices in under half a second—often under 0.3 seconds—at $0.015 per 1,000 characters, making natural phone-based customer service bots much cheaper and faster.
Google DeepMind just unveiled Genie, an 11-billion-parameter foundation world model that conjures playable 2D platformers from a single image or sketch. Trained on over 200,000 hours of unsupervised gameplay videos, it automatically infers character controls and physics. Why it matters: It turns static art into interactive worlds and could serve as an endless simulator to train future autonomous AI agents.
OpenAI is reportedly working on an autonomous AI agent capable of taking control of user devices to handle tasks directly, according to The Information. Instead of just answering prompts, it could click, type, and transfer data across apps for you. Why it matters: bringing JARVIS-style automation to everyday devices sounds incredible, but handing full device access to cloud AI creates serious privacy and security hurdles.
The Browser Company just revealed its AI-driven roadmap for the Arc desktop browser. Features like Instant Links skip search engines entirely by opening relevant tabs directly from your query, while Live Folders automatically pulls in live feeds. It is also prepping Arc Explore, bringing automated AI web research to the desktop.
OpenAI just rolled out a feature allowing paid ChatGPT users to tag in custom GPTs into any conversation by simply typing "@". The summoned GPT inherits the full chat context, making it easy to seamlessly chain different tools together. OpenAI is looking to boost adoption for its GPT Store, where custom GPTs currently drive under 3% of web traffic.
A technical blog post just dissected the real-world intersections of crypto and AI, pointing out where the hype meets practical utility. The most promising applications involve AIs acting as players in micro-prediction markets or smart contract assistants, but using models to enforce governance rules exposes them to severe adversarial attacks. Why it matters: Decentralization and open models clash with machine learning vulnerabilities, meaning not all crypto-AI mashups are built to last.
OpenAI just confirmed in an email to developers that its GPT Store will finally launch next week after being delayed by November's leadership chaos. The platform will let creators share, discover, and monetize custom bots through a usage-based revenue-sharing model. It could be the long-awaited App Store moment for AI agents.
Robotics researchers just introduced AutoRT, SARA-RT, and RT-Trajectory to make robots faster, safer, and better at generalization. AutoRT uses foundation models to safely orchestrate fleets and gather training data, SARA-RT slashes Transformer complexity to boost speed by 14%, and RT-Trajectory doubles task success rates on unseen scenarios.
Anthropic just updated its commercial terms to defend customers against copyright infringement claims, promising to cover settlements and judgments from authorized use. The company also introduced a beta Messages API that streamlines prompt construction and sets the stage for upcoming function calling.
OpenAI just delayed the launch of its GPT Store until next year after a chaotic couple of weeks dealing with Sam Altman's ouster and reinstatement. The marketplace will allow users to share and monetize custom chatbots, but the company says unexpected events kept it busy. In the meantime, OpenAI has rolled out a few usability updates to the custom GPT builder.
OpenAI just rolled out custom GPTs to subscribers after beating back a coordinated DDoS attack that took down ChatGPT and its developer API. The rollout was stalled after traffic surges—initially blamed on DevDay hype—turned out to be foul play claimed by Anonymous Sudan. Now the service is stabilized and custom bots are live.
