From TIME · Billy Perrigo · · 1m 30s
AI companies race toward self-improving systems
A new report reveals how Anthropic, OpenAI and others are automating AI development itself, with insiders predicting the breakthrough could arrive within months.
In brief
A new report reveals how Anthropic, OpenAI and others are automating AI development itself, with insiders predicting the breakthrough could arrive within months. Anthropic and OpenAI are actively pursuing recursive self-improvement, where AI systems improve themselves automatically. A new Gladstone AI report based on…
The tipping point nobody talks about
Anthropic and OpenAI have warned that AI may soon improve itself automatically, a process called recursive self-improvement. This could accelerate AI capabilities dramatically but might also mean developers lose control entirely.
A rare glimpse inside the labs
Gladstone AI published a report Wednesday based on conversations with half a dozen researchers inside top AI companies.
The timeline is measured in months
Most insiders Gladstone spoke to believe recursive self-improvement will start within the next three to twelve months, and possibly sooner, assuming there is no U.S. government intervention or slowdown.
“believe RSI will start within the next 3 to 12 months, and possibly sooner, assuming there's no U.S. government intervention or slowdown”
How AI development could automate itself
AI companies are actively working to automate all parts of the development pipeline: generating new ideas, running experiments to test them, and interpreting results.
Computing power becomes the bottleneck
Companies with more computing power can test more ideas and reach recursive self-improvement first. The rarest ideas are those that perform better as more computing power is added, even if they look like failures at…
Regulators need to see the process
Jeremie Harris, Gladstone's CEO, says the report aims to help lawmakers understand recursive self-improvement so they can regulate it.
“a human needs to be there”
The short version
Anthropic and OpenAI are actively pursuing recursive self-improvement, where AI systems improve themselves automatically. A new Gladstone AI report based on insider conversations suggests this could happen within three to twelve months.
