Chinese AI startup z.ai made headlines this week with the release of its new large language model, GLM-5, which achieved a record-low hallucination rate, according to VentureBeat. The model, which is open source, also leverages a new reinforcement learning technique called "slime." Meanwhile, researchers at MIT, the Improbable AI Lab, and ETH Zurich developed a new fine-tuning method that allows LLMs to learn new skills without forgetting old ones, as reported by VentureBeat.
GLM-5, the latest in z.ai's GLM series, retains an open-source MIT License, making it suitable for enterprise deployment. It achieved a score of -1 on the AA-Omniscience Index, representing a 35-point improvement over its predecessor. This places GLM-5 at the forefront of the AI industry in knowledge reliability, surpassing U.S. competitors like Google, OpenAI, and Anthropic, by knowing when to abstain rather than fabricate information, according to VentureBeat.
The MIT researchers' new technique, called self-distillation fine-tuning (SDFT), allows models to learn directly from demonstrations and their own experiments by leveraging the inherent in-context learning abilities of modern LLMs. Experiments show that SDFT consistently outperforms traditional supervised fine-tuning while addressing the limitations of reinforcement learning, according to VentureBeat.
While advancements in AI continue, concerns persist regarding its potential misuse. AI is already making online crimes easier, and the situation could worsen, according to MIT Technology Review. Hackers are using AI tools to reduce the time and effort required to orchestrate attacks, lowering the barriers for less experienced attackers. Some in Silicon Valley warn that AI is on the brink of being able to carry out fully automated attacks. Security researchers argue that the immediate risks posed by AI, which is already speeding up and increasing the volume of scams, should be the primary focus.
In related news, a recent study in Nature Energy found that EVs from scooters to minibuses could be cheaper to own than gas-powered vehicles in Africa by 2040, according to MIT Technology Review. However, the technology still faces major challenges in some African markets, including limited grid and charging infrastructure.
Discussion
AI Experts & Community
Be the first to comment