All articles tagged with #self improving ai
A senior Anthropic safety researcher warns there is more than a 10% chance AI could kill all humans by the end of the decade, just as a colleague resigns over lax safety; the piece notes the industry’s rush to self-improving systems with no clear safety plan, while safety concerns mount ahead of anticipated IPOs.
Meta's researchers report observing their AI systems improving themselves without human input, marking a step toward artificial superintelligence (ASI), which could surpass human intelligence and lead to an exponential 'intelligence explosion.' CEO Mark Zuckerberg emphasizes caution in releasing such powerful models to the public, highlighting the potential for transformative progress and personal empowerment through superintelligence.
DeepMind has introduced a self-improving AI model called RoboCat that can self-generate new training data to improve its technique without too much human interference. RoboCat learns much faster than other state-of-the-art models and can pick up a new task with as few as 100 demonstrations. This reduces the need for human-supervised training and is an important step towards creating a general-purpose robot.
Google's DeepMind has developed an AI agent called RoboCat that can learn to operate different robotic arms and take on different tasks with as little as 100 examples. RoboCat can create its own training data to help sharpen performance and learn new skills within a few hours. The self-improving AI can learn multiple tasks on multiple robotic arms, potentially opening the door to general-purpose robots that could lead to more flexible automation and decreased costs.