Ethan Mollick said he tested an early version of Anthropic's most advanced AI model Claude Fable and found it both remarkable and unsettling to use. Claude Fable Tested Mollick, a professor a
Ethan Mollick said he tested an early version of Anthropic's most advanced AI model Claude Fable and found it both remarkable and unsettling to use.
Mollick, a professor at the Wharton School of the University of Pennsylvania, wrote in an essay Tuesday that he tested Claude 5 Fable early. The system is the first Mythos-class model Anthropic has released publicly. He said it beat every public model he had tried by a wide margin.
From a single prompt and one piece of feedback, it produced what Mollick called the most sophisticated academic social science paper he had seen from an AI. It also wrote a 10-page rhyming poem about a haircut, with every word starting with the letter s.
To build a travel-time map, the model launched cheaper Claude Sonnet agents and gathered data on over 2,200 flights. A separate request produced research software after nine and a half hours. It cannot make images, so it built every chart and object using math, while its guardrails block security tasks.
Also Read:OpenAI Joins $852B IPO Race As Anthropic And SpaceX Charge Ahead
Mollick said the work felt delightful and unnerving, because he asked for something and it happened. The model followed his instructions closely, he wrote, but revealed little of the hundreds of small choices it made.
He noted that Fable costs about twice as much as Anthropic's Opus model. Its guardrails also fall back to a weaker system too often, he wrote. He compared his role to a client signing off on a studio's finished work.
A year ago, Mollick compared working with these tools to casting a spell as a wizard. The spell has grown powerful enough, he wrote, that he no longer feels like the one casting it.
Read Next:Bitcoin Drawdown Reveals Why Big Buyers Still Want More BTC
