Google releases its own ‘reasoning’ AI model

Google has launched a new AI model called Gemini 2.0 Flash Thinking Experimental, which is in its experimental stages and available through the AI Studio platform for model prototyping. The model aims to enhance multimodal understanding, logical reasoning, and solving complex problems in fields like programming, mathematics, and physics. It is characterized by its ability to self-reason and validate its conclusions, though it suffers from relatively slow response times.

The model employs a "reasoning during inference" technique, pausing before responding to analyze additional related inputs and providing explanations for its results. However, early tests revealed some shortcomings, such as incorrectly counting letters in words.

The AI industry is experiencing a surge in the development of reasoning models, such as OpenAI's o1 and others from companies like Alibaba and DeepSeek. This trend is driven by the declining effectiveness of traditional scaling techniques for improving generative models.

Nevertheless, these models face challenges, including high costs and difficulties in maintaining progress rates, raising questions about the feasibility of this approach in AI development.

Read the full article