Why current AI models won't make scientific breakthroughs, according to a top tech exec

Hugging Face co-founder Thomas Wolf asserts that current AI models, including those from OpenAI, are unlikely to achieve major scientific breakthroughs on a Nobel Prize level, citing their tendency to align with user prompts and predict the most probable next token rather than generating contrarian, novel truths. This view contrasts with more optimistic outlooks from figures like Sam Altman and Dario Amodei, with Wolf suggesting AI's more realistic role is as a "co-pilot" for scientists, aiding research and idea generation, as exemplified by tools like Google DeepMind's AlphaFold.

Analysis

The discourse surrounding artificial intelligence's potential is being tempered by key industry figures, creating a more nuanced investment landscape. Thomas Wolf, co-founder of the $4.5 billion startup Hugging Face, posits that current large language models are structurally unsuited for generating Nobel-level scientific breakthroughs. His core argument is that these models are engineered to predict the most probable 'next token' and align with user prompts, a function antithetical to the contrarian thinking that underpins major scientific discoveries. This perspective directly challenges the more bullish outlooks from leaders at OpenAI and Anthropic, with the latter's CEO having projected a compression of 50-100 years of biological progress into the next 5-10 years. The article suggests AI's immediate, practical role is as a 'co-pilot for a scientist,' enhancing research productivity rather than supplanting human ingenuity. Google DeepMind's AlphaFold, which assists in analyzing protein structures, is cited as a prime example of this successful co-pilot model, contributing to the positive sentiment score (0.5) for Alphabet (GOOGL). The emergence of new startups like Lila Sciences and FutureHouse indicates that while current models may be limited, the private markets are actively funding efforts to transcend these limitations, highlighting a key area of future development and potential disruption.

Key Decisions for Investors

• Investors should differentiate between the hype of AI-driven scientific revolution and the current reality of AI as a productivity tool; near-term value is more likely to be found in companies deploying AI as a 'co-pilot' to enhance existing workflows.
• The positive application of Google's AlphaFold suggests focusing on companies with tangible, proven AI applications that assist human experts, as these present more immediate and verifiable ROI than speculative, long-term bets on AI achieving autonomous discovery.
• Given the debate on AI's limitations and the rise of new startups aiming to overcome them, it is prudent to monitor the venture capital space and R&D in next-generation AI architectures for early signs of a technological shift that could unlock true breakthrough potential.