© ROOT-NATION.com - Use of content is permitted with a backlink.
Anthropic CEO Dario Amodei said that artificial intelligence (AI) could surpass “almost all humans at almost everything” shortly after 2027.
While the capabilities of artificial intelligence are indeed evolving, such rapid progress may seem at odds with data showing that AI still fails in over 95% of remote freelance projects, and continues to struggle with hallucinations, long-term planning, and forms of abstract thinking that humans come easily to master.
However, recent research from MIT has found evidence that large language models (LLMs) can rapidly acquire new abilities, going from near-total failure to near-total success in just a few years. If this extends to the entire economy, workers could be caught off guard by AI’s progress.
In their study, MIT researchers call such sudden leaps in ability “surge waves” and consider whether they are an economy-wide phenomenon or whether AI progress is gradual, like a “rising tide.” By analyzing thousands of real-world tasks, the team concluded that while AI capabilities are improving rapidly, they are developing more smoothly, suggesting that “spikes” are the exception rather than the rule.
AI productivity is increasing evenly across industries and for tasks of varying durations, indicating a gradual increase in ability. “This doesn’t guarantee protection for workers, the tide may still be rising quickly, but it does mean that workers and policymakers who track progress will be able to see AI improvements in advance,” explains senior author Neil Thompson.
The possibilities of artificial intelligence are already significant. The researchers focused on 63% of the tasks in the US economy that are text-based and thus potentially performed by LLMs. Among these tasks, given the right information, LLMs were able to perform 60% of the tasks at a level that a supervisor would consider “minimally adequate” without human intervention. Only 26% of tasks were performed at a “superior” level. Lead researcher Matthias Mertens said: “LLMs demonstrated impressive competence even on their own.”
AI abilities are developing rapidly. Although the MIT study believes that 2027 is too optimistic for AI to completely surpass humans, it confirms rapid progress. Forecasts show that by 2029, AI will be able to succeed in 80% of tasks. However, according to Dr. Thompson, “this depends on further development of AI hardware and algorithms and model scaling. If progress slows down, the rate of growth of AI abilities will also slow down.”
The findings of the study are important for policymakers and businesses seeking to prepare for the changes that will bring new opportunities and challenges associated with artificial intelligence.
Read also:
- Ukrainians are actively buying electric cars: What happened
- Record after record: How Ukraine is taking over esports broadcasts