Not to be outdone by OpenAI, Google releases its own “reasoning” AI model
Dec 20, 2024 - arstechnica.com
Google DeepMind's chief scientist, Jeff Dean, highlights the potential of increased computing power for AI models, noting promising results when inference time computation is enhanced. The model operates by pausing to evaluate multiple related prompts to deliver the most accurate answer. Since OpenAI's entry into the "reasoning" field with o1-preview and o1-mini, companies like DeepSeek and Alibaba have launched their own reasoning models, DeepSeek-R1 and QwQ, respectively. These models aim to solve complex problems, but questions about their accuracy and usefulness persist, alongside concerns over high computing costs, exemplified by OpenAI's $200 monthly fee for ChatGPT Pro.
Despite these challenges, Google is committed to advancing this AI technique. Logan Kilpatrick from Google's AI Studio described it as "the first step in our reasoning journey" in a post on X. The pursuit of reasoning models reflects a broader industry trend, though their long-term viability remains uncertain due to the significant computational resources required.
Key takeaways:
Google DeepMind's model benefits from increased computing power, leading to promising results in reasoning tasks.
Several companies, including DeepSeek and Alibaba, are developing their own reasoning models to compete with OpenAI's advancements.
Reasoning models show potential in solving complex problems but face challenges regarding their usefulness, accuracy, and high computing costs.
Google is committed to advancing reasoning AI techniques, viewing it as an important step in their AI development journey.
