The artificial intelligence (AI) industry is witnessing a surge in the development of reasoning models, with multiple labs releasing their own versions in quick succession. This trend is largely attributed to the release of OpenAI's o1, a so-called reasoning model, which has sparked a competitive frenzy among AI companies to refine generative AI technology.

DeepSeek, an AI research company funded by quantitative traders, launched a preview of its first reasoning algorithm, DeepSeek-R1, in early November. Alibaba's Qwen team also unveiled what it claims is the first "open" challenger to o1. This explosion of reasoning models is driven by the search for novel approaches to refine generative AI tech, as traditional "brute force" techniques to scale up models are no longer yielding the improvements they once did.

The global AI market, which reached $196.63 billion in 2023 and is expected to be worth $1.81 trillion by 2030, is fueling the competitive pressure on AI companies to maintain the current pace of innovation. OpenAI has claimed that reasoning models can "solve harder problems" than previous models and represent a step change in generative AI development. However, not everyone is convinced that reasoning models are the best path forward.

Ameet Talwalkar, an associate professor of machine learning at Carnegie Mellon, expressed caution, stating that he would "question the motives" of anyone claiming with certainty that they know how far reasoning models will take the industry. Talwalkar warned that AI companies have financial incentives to offer rosy projections about the capabilities of future versions of their technology, and that the broader AI research community should avoid blindly believing the hype and marketing efforts of these companies.

Two significant downsides of reasoning models are their expensiveness and power-hungry nature. OpenAI's API, for instance, charges $15 for every ~750,000 words o1 analyzes and $60 for every ~750,000 words the model generates, which is between 3x and 4x the cost of OpenAI's latest "non-reasoning" model, GPT-4o. The company has also introduced a more advanced o1 tier, o1 pro mode, that costs an eye-watering $2,400 a year.

Reasoning models require a lot of computing resources to run, as they attempt to check their own work as they operate. This helps them avoid some pitfalls that normally trip up models, but it also means they often take longer to arrive at solutions. OpenAI envisions future reasoning models "thinking" for hours, days, or even weeks on end, with the potential payoffs being breakthroughs in areas like battery technology and cancer drugs.

However, the value proposition of today's reasoning models is less obvious. Costa Huang, a researcher and machine learning engineer at the nonprofit org Ai2, notes that o1 isn't a very reliable calculator, and social media searches reveal a number of o1 pro mode errors. Guy Van Den Broeck, a professor of computer science at UCLA, asserts that reasoning models aren't performing actual reasoning and are thus limited in the types of tasks that they can successfully tackle.

Despite these limitations, it's likely that reasoning models will continue to improve with time, driven by the strong market incentive to boost their capabilities. However, Talwalkar worries that big labs will gatekeep these improvements, hindering the research community's ability to engage with these ideas. As more people work on this direction, he expects reasoning models to quickly advance, but believes that most models will be offered by large industrial labs like OpenAI.

The surge in reasoning model development is a significant trend in the AI industry, with far-reaching implications for the future of artificial intelligence. As the technology continues to evolve, it will be crucial to monitor its progress and potential applications, while also addressing the limitations and concerns raised by experts in the field.