Google has launched its latest AI model, Gemini 2.0 Flash Thinking Experimental, aiming to outdo OpenAI with advanced reasoning techniques. The experimental model uses runtime “reasoning” methods similar to o1 from OpenAI and achieves deeper insights on problems fed into it. Built on the Google AI Studio platform, the model runs on top of its newly released Gemini 2.0 Flash. However, early tests revealed accuracy issues in basic tasks, such as miscounting the number of R’s in “strawberry”. The reasoning models incorporate self-checking mechanisms and feedback loops, requiring more computing time and potentially slower response times. This approach is seen as a scaling method to overcome diminishing returns on traditional training methods.
Source: https://arstechnica.com/information-technology/2024/12/not-to-be-outdone-by-openai-google-releases-its-own-reasoning-ai-model