Google has revealed a new artificial intelligence (AI) model referred to as “reasoning” AI, but it remains in the experimental phase. This model, named Gemini 2.0 Flash Thinking Experimental, is accessible through AI Studio, which is Google’s platform for AI prototyping. The model card indicates that it excels in multimodal understanding, reasoning, and coding, claiming the capability to tackle the most complex problems in areas such as programming, mathematics, and physics.
How is Gemini 2.0 different from previous models?
In a recent post on X, Jeff Dean, chief scientist at Google DeepMind, highlighted that this model is “trained to use thoughts to strengthen its reasoning.” He also pointed out that it benefits from the enhanced speed associated with the faster Gemini Flash 2.0 model.
A demonstration shared by Dean illustrates how Gemini 2.0 Flash Thinking approaches a physics problem by “thinking” through a series of steps prior to presenting a solution.
This process doesn’t replicate human reasoning precisely; rather, it enables the machine to decompose instructions into smaller, manageable tasks that can lead to stronger outcomes.
What examples showcase Gemini 2.0’s capabilities?
Another illustration, shared by Logan Kilpatrick, product lead at Google, shows the model reasoning through a problem that incorporates both visual and textual components. Kilpatrick remarked, “This is just the first step in our reasoning journey.” Users are invited to experiment with Gemini 2.0 Flash Thinking on Google’s AI Studio.
In a post on X, Logan Kilpatrick described Gemini 2.0 Flash Thinking Experimental as “the first step in [Google’s] reasoning journey.” Jeff Dean reiterated in his own post that the model is “trained to use thoughts to strengthen its reasoning.” He noted, “We see promising results when we increase inference time computation,” which pertains to the computing resources utilized to “run” the model while it contemplates a question.
How does Gemini 2.0 compare with other reasoning models?
Gemini 2.0 Flash Thinking Experimental builds upon Google’s recently announced Gemini 2.0 Flash model and appears to share design similarities with OpenAI’s o1 and other so-called reasoning models. Unlike many AI systems, reasoning models possess the ability to effectively fact-check themselves, which helps them sidestep some common pitfalls that often hinder AI models. However, a significant drawback is that reasoning models generally require more time—typically seconds to minutes longer—to arrive at solutions.
What is the response time of Gemini 2.0 Flash Thinking?
When given a prompt, Gemini 2.0 Flash Thinking Experimental pauses before generating a response, deliberating over various related prompts and “explaining” its reasoning throughout the process. Eventually, the model summarizes what it deems to be the most accurate answer.
Read more | How to access ChatGPT on WhatsApp? All you need to know
Reasoning models gaining popularity
Following the introduction of o1, there has been a surge of reasoning models emerging from competing AI laboratories—not solely from Google. In early November 2024, DeepSeek, an AI research firm backed by quantitative traders, debuted a preview of its first reasoning model, named DeepSeek-R1. That same month, Alibaba’s Qwen team revealed what it claimed to be the first “open” competitor to o1.
What has sparked this influx of reasoning models? Primarily, it stems from the pursuit of innovative methods to enhance generative AI. Nonetheless, skepticism exists regarding whether reasoning models represent the optimal path forward. They often incur significant costs due to the extensive computational power required for their operation. While they have shown strong performance on benchmarks thus far, it remains uncertain if reasoning models can sustain this momentum.
Recently, there have been several notable advancements in the AI sector, including Google’s unveiling of the upgraded Gemini 2.0 model earlier this month, which is part of the company’s initiative toward “agentic” AI. In parallel, OpenAI made the complete version of its o1 reasoning model available to ChatGPT subscribers.
OpenAI has announced a new feature that brings ChatGPT to WhatsApp, enhancing accessibility for its vast user base. With over 2.7 billion users globally, this integration allows individuals to engage in seamless conversations with the AI tool directly through the messaging platform. To initiate a chat, users can easily add the number +1 1800 242 8478 to their contacts. This feature enables immediate interaction with ChatGPT without the requirement of logging into an OpenAI account, streamlining the experience for users looking to harness the capabilities of AI on a familiar platform.
