News
1. AI for scientific discovery
An international team of scientists has launched Polymathic AI, a new research collaboration that will use AI models trained on diverse scientific data to uncover connections and aid discovery across physics, astrophysics, and eventually other fields. The project aims to create a publicly available foundation model that researchers worldwide can build on to improve analyses in their domain (TechXplore).
2. A multimodal architecture for AI agents
Adept AI is open-sourcing Fuyu-8B, a small multimodal model that can understand images, charts, diagrams, and documents. The model performs well on standard benchmarks despite optimization for the company’s internal use cases (Adept AI).
3. ERNIE 4.0 is claimed to be in the same league as GPT-4
Baidu unveiled an upgraded version of its AI chatbot ERNIE 4.0, claiming it can match GPT-4, while also announcing integration of the technology into its search engine, maps, and other products (CNN).
4. A new drug for COVID-induced smell loss
New clinical trial data show the antiviral pill ensitrelvir can shorten the duration of loss of smell and taste in people with mild to moderate COVID-19 (Nature).
5. Living guidelines for the use of generative AI in research
Some propose creating an independent scientific body to audit AI systems for safety and effectiveness, develop living guidelines for use of AI in research, and certify tools, with oversight by an expert committee (Nature).
6. A new IBM chip for AI
IBM has developed a new computer chip called NorthPole that integrates memory and processing power, eliminating the need to access external memory and overcoming a major speed and efficiency bottleneck in AI computing (Nature).
Articles
1. How ‘A.I. Agents’ That Roam the Internet Could One Day Replace Workers (NYT)
Researchers are turning chatbots like ChatGPT into AI agents that can use software tools and websites, playing games, scheduling meetings, and building charts autonomously. While promising to make office workers more efficient, these AI agents could eventually automate almost any white-collar job and have huge economic consequences.
2. Elon Musk wants to merge humans with AI. How many brains will be damaged along the way? (Vox)
Neuralink’s brain implants could help paralyzed people control devices with their thoughts, but former employees allege the company uses an unnecessarily invasive approach that damages animal brains. There are ethical concerns about mental privacy and authoritarian surveillance, as the technology may enable unprecedented access to people’s thoughts.
3. Minds of machines: The great AI consciousness conundrum (MIT Tech Review)
Experts are divided on whether advanced AI like ChatGPT could become conscious, sparking intense ethical debates on how to detect and protect a potentially suffering AI while avoiding risks of incorrectly attributing consciousness.
