According to Apple researchers, the race to prepare the Artificial General Intelligence (AGI) is still a long journey, which has learned that leading AI’s leading models still have problems in reasoning.
Recent updates of AI’s large language models (LLM) such as Open Chattagpat and Anthropic Cloud include large reasoning models (LRM), but their basic abilities, scaling features and limits are “insufficiently understood”, “a dissertation called” Mutual Mutual “.
He noted that the current diagnosis is primarily focused on math and coding benchmarks, “emphasizing the correct response accuracy.”
However, this diagnosis does not provide insights about the reasoning capabilities of AI models.
Research contradicts the expectation that artificial general intelligence remains only a few years.
Apple researchers test AI models “thinking”
Researchers developed various puzzle games to test different forms of “thinking” and “non-thinking”, beyond the standard mathematics standards of Claude Swant, Openi’s O3-Mini and O1, and V3 chat boats.
He discovered that “Frontier LRMS faces a complete accuracy beyond some complications,” do not make the reasoning effectively, and contrary to the expectations of Agi capabilities, their edge disappears with increasing complexity.
“We have found that there are limits to the LRM’s exact count: they fail to use clear algorithms and arguments in contradictory puzzles.”
Researchers say AI chat boats are thinking more
He found and observed contradictory and shallow arguments with the models, AI chat boats quickly created the right answers and then wandering in the wrong reasoning.
Related: The role of stabilizing AI in Web 3, challenging defect and gaming: Daprader
Researchers concluded that the LRMS was imitating samples of reasoning without making them truly internal or generalized, which is less than the AGI level argument.
“This insight challenges the existing assumptions about the capabilities of the LRM and suggests that the current approach may face the main obstacles to the general reasoning.”
AGI’s development race
AGI AI is the sacred stone of development, a state where a machine can think and argue like a human being and is equal to human intelligence.
In January, Open CEO, Sam Altman, said the firm was near the construction of AGI more than ever before. He said at the time, “We are now confident that we know how to build AGI because we have traditionally understood it.”
In November, Antarropic CEO Dario Amody said that the AGI will exceed human capacity in the next two years. “If you only go to the rate that is increasing, it forces you to think that we will reach 2026 or 2027,” he said.
Magazine: Ignore AI jobs, Domers, AI job is good for Good PWC says: AIII
