A few days ago,Anthropic Core Technology member Sholto Douglas was interviewed by Unsupervised Learning, where he mentioned "TheAI I have a variety of opinions on how to work for and live with humans in the future," he said.
One of the most striking aspects of the interview was Douglas' prediction of AI competence in human jobs: by 2027-2028, AI models will have the ability to automate virtually all white-collar jobs. He further explains that this progress is not science fiction, but is based on the rapid improvements in task complexity and sustained action capabilities of current models.
At the same time, Douglas also talked about one of the biggest parts of AI's ability to do a human job well - theAI AgentThe core obstacle for Agents, it argues, is 'reliability'. According to him, the core obstacle for Agents is 'reliability'. It said that although the current model is not 100% reliable, all the data and trends are proving that human beings are moving towards the goal of 'expert level' or even 'super-human level'.
It's worth noting that Douglas also mentioned the impact of AI on GDP in the interview. He predicts that the initial impact of AI on global GDP could be comparable to the reshaping of the world economy by the 'rise of China', and at a much faster rate.