Tao Lin, an AI safety researcher at Anthropic, stated in an interview that by the end of 2027, artificial intelligence will surpass top experts in nearly all fields, and he estimates the probability of humanity losing control over its future at approximately 20%.

Forecast on Automatic Model Improvement and Replacement of Human Labor

According to Lin, within a year AI will be able to independently create smarter versions of itself, after which technological development may proceed without human involvement. He predicts a widespread replacement of human labor: most people will be unemployed and rely on government benefits, while physical work will require "billions of robots."

Risks of Using Claude and Requests for Military Applications

Lin explained that Claude is protected against dangerous uses and that Anthropic receives requests to create military drones, including from parties involved in the Russia-Ukraine conflict; the company strives to block such requests.

Need to Slow Down Development and Developers’ Position

The researcher noted that developers should slow down progress by a factor of four due to intense competition and reduced safety checks. He also stated that developers are aware of the risk of humanity’s destruction but continue working and creating more powerful models.

The full interview is available in video format: Tao Lin interview on YouTube (with Russian subtitles and an English version).