The Rise of Reasoning Models in AI
The meteoric rise of DeepSeek R-1 has put the spotlight on an emerging type of AI model called a reasoning model. As generative AI applications move beyond conversational interfaces, reasoning models are likely to grow in capability and use, which is why they should be on your AI radar.
What are Reasoning Models?
Reasoning models are a type of large language model (LLM) that can perform complex reasoning tasks. Unlike LLMs, which quickly generate output based solely on a statistical guess of what the next word should be, reasoning models take time to break a question down into individual steps and work through a "chain of thought" process to come up with a more accurate answer. In this manner, a reasoning model is much more human-like in its approach.
How Do Reasoning Models Work?
According to OpenAI, reasoning models work through problems more like a human would compared to earlier language models. They involve a chain-of-thought process that includes additional tokens. This process is similar to how a human may think for a long time before responding to a difficult question. The model learns to hone its chain of thought and refine the strategies it uses, recognize and correct its mistakes, break down tricky steps into simpler ones, and try a different approach when the current one isn’t working.
The Rise of DeepSeek R-1
The introduction of DeepSeek R-1 was a breakthrough, as it dramatically reduced computational requirements. The company behind DeepSeek claims that it trained its V-3 model on a small cluster of older GPUs that only cost $5.5 million, much less than the hundreds of millions it reportedly cost to train OpenAI’s latest GPT-4 model. And at $0.55 per million input tokens, DeepSeek R-1 is about half the cost of OpenAI o3-mini.
The Impact on AI Research and Development
The success of DeepSeek R-1 is forcing AI researchers to rethink their approach to developing and scaling AI. Instead of racing to build ever-bigger LLMs that sport trillions of parameters and are trained on huge amounts of data culled from a variety of sources, the success of reasoning models like DeepSeek R-1 suggest that having a larger number of smaller models trained using a mixture of experts (MoE) architecture may be a better approach.
Conclusion
The rise of reasoning models is a significant development in the field of AI. As the pace of AI evolution continues to accelerate, it is likely that these types of surprises and shocks will become more frequent. This may make for a bumpy ride, but it will ultimately create AI that is more capable and useful, and that’s ultimately a good thing for all of us.
FAQs
Q: What is a reasoning model?
A: A reasoning model is a type of large language model (LLM) that can perform complex reasoning tasks.
Q: How do reasoning models work?
A: Reasoning models work through problems more like a human would, involving a chain-of-thought process that includes additional tokens.
Q: What is the impact of the rise of DeepSeek R-1 on AI research and development?
A: The success of DeepSeek R-1 is forcing AI researchers to rethink their approach to developing and scaling AI, suggesting that smaller models trained using a mixture of experts (MoE) architecture may be a better approach.
Q: What is the cost of training a reasoning model?
A: The cost of training a reasoning model can vary, but DeepSeek R-1 was trained on a small cluster of older GPUs that cost $5.5 million, while OpenAI’s o3-mini was trained on hundreds of millions of dollars.
