An AI safety researcher at Anthropic resigned this week with a stark warning that the "world is in peril," citing concerns about artificial intelligence, bioweapons, and the state of the world. Mrinank Sharma, who shared his resignation letter on X, announced he would be leaving the US firm to pursue writing, study poetry, and relocate to the UK. This news comes amid a flurry of activity and concern in the AI field, including advancements in large language models and the development of tools to aid those with debilitating illnesses.
Sharma's departure from Anthropic, known for its Claude chatbot, follows a similar resignation from an OpenAI researcher who expressed worries about the company's decision to introduce advertisements into its chatbot. Sharma's letter did not provide specific details about his concerns, but his decision to leave the field entirely and "become invisible" underscores the growing anxieties surrounding the rapid development and potential risks of AI.
Meanwhile, advancements in AI continue to emerge. Researchers at Nvidia have developed a technique called dynamic memory sparsification (DMS) that can reduce the memory costs of large language model reasoning by up to eight times, according to VentureBeat. This method compresses the key value (KV) cache, the temporary memory LLMs use, without significantly degrading the model's performance. Experiments show that DMS allows LLMs to "think" longer and explore more solutions without the need for additional resources.
In other news, AI is also being used to help those with debilitating illnesses. A musician who lost his voice to amyotrophic lateral sclerosis (ALS) was able to sing again with the help of AI. The musician, Patrick Darling, performed on stage with his bandmates for the first time in two years, according to MIT Technology Review. His performance was made possible by AI technology that recreated his voice, allowing him to share his music with his audience once more.
Further developments in the field include the ability to spin up cloud sandboxes, run commands, transfer files, and automate browsers using Claude Code, Codex, and other agents, according to Hacker News. This functionality, provided by cloudrouter, gives coding agents the ability to create virtual machines and utilize GPUs from the command line.
Finally, research continues into the origins of life. According to Ars Technica, scientists are exploring the potential of RNA molecules to self-replicate. This research focuses on the ability of RNA to carry genetic information and fold into three-dimensional structures that act as catalysts. This could potentially help scientists understand how early life may have functioned.
Discussion
AI Experts & Community
Be the first to comment