AI models still far from AGI-level reasoning: Apple researchers

Apple researchers have published a paper questioning the reasoning capabilities of leading large language models (LLMs) like ChatGPT and Claude, finding that their performance collapses beyond certain complexities and that they struggle to generalize reasoning effectively. The study, titled "The Illusion of Thinking," challenges the prevailing assumption that artificial general intelligence (AGI) is imminent, suggesting that current approaches to LLMs face fundamental barriers to achieving human-like reasoning, despite claims from OpenAI and Anthropic that AGI is only a few years away.

Analysis

Apple researchers have presented findings in a June paper, "The Illusion of Thinking," which temper expectations for rapid advancements towards artificial general intelligence (AGI) by current leading large language models (LLMs) such as OpenAI's ChatGPT and Anthropic's Claude. The research indicates that these models, despite incorporating large reasoning models (LRMs), exhibit a "complete accuracy collapse beyond certain complexities" and fail to generalize reasoning effectively, with their performance edge diminishing as task complexity increases. This conclusion, based on tests using puzzle games designed to probe beyond standard mathematical and coding benchmarks, suggests that current LLMs tend to mimic reasoning patterns rather than internalizing them, and can "overthink," generating correct answers initially before devolving into incorrect reasoning. These findings challenge the optimistic AGI timelines projected by industry leaders, such as OpenAI's Sam Altman and Anthropic's Dario Amodei who anticipated AGI within a few years, and suggest that prevailing approaches "may be encountering fundamental barriers to generalizable reasoning." The overall sentiment surrounding these findings is moderately negative (-0.65 score), reflecting a cautious outlook on the current trajectory of AGI development, although Apple's (AAPL) specific involvement and research publication is viewed with a neutral to slightly positive sentiment (0.2 ticker score), potentially indicating market perception of a rigorous and realistic approach from the company.

Key Decisions for Investors

• Investors should moderate expectations for near-term AGI realization based on current LLM capabilities, as highlighted by Apple's research suggesting significant reasoning limitations.
• Scrutinize valuations of AI companies heavily dependent on rapid AGI progress via existing LLM pathways, considering the potential for "fundamental barriers" and the cautious market sentiment indicated by the report.
• Investors may consider Apple's research stance as a differentiator, potentially signaling a more grounded, long-term AI strategy that could be less susceptible to hype cycles compared to entities making more aggressive AGI claims.
• Monitor subsequent research on AI reasoning capabilities and alternative development pathways, as these will be critical indicators for the AI sector's trajectory and investment viability.