Google has released what it calls a brand new “logical” AI model – but it surely's still within the experimental stage and our transient testing shows there's actually room for improvement.
The latest model, called Gemini 2.0 Flash Thinking Experimental (a mouthful, after all), is obtainable in AI Studio, Google's AI prototyping platform. A model map describes it as “best fitted to multimodal understanding, pondering and coding” with the power to “reason about probably the most complex problems” in areas resembling programming, mathematics and physics.
In one post Logan Kilpatrick, product lead for AI Studio, called Gemini 2.0 Flash Thinking Experimental based on Jeff Dean, chief scientist of Google DeepMind, Google's AI research division, said in his own post that Gemini 2.0 Flash Thinking Experimental is “trained to make use of thoughts to strengthen his argument.”
“We are seeing promising results as we increase the inference computation time,” Dean said, referring to the quantity of computation required to “run” the model when testing an issue.
It's still an early version, but take a look at how the model tackles a difficult puzzle using visual and textual clues: (2/3) pic.twitter.com/JltHeK7Fo7
– Logan Kilpatrick (@OfficialLoganK) December 19, 2024
Gemini 2.0 Flash Thinking Experimental relies on Google's recently announced Gemini 2.0 Flash model and appears to be similar in design to OpenAI's o1 and other so-called reasoning models. Unlike most AI models, reasoning models effectively self-check the facts, allowing them to avoid a few of the pitfalls that typically trip up AI models.
One drawback is that argumentation models often take longer – normally seconds to minutes longer – to reach at solutions.
When prompted, Gemini 2.0 Flash Thinking Experimental pauses before responding, considers a series of related prompts, and “explains” the reasoning because it does so. After some time, the model summarizes what it thinks is probably the most accurate answer.
Well – that’s exactly what’s speculated to occur. When I asked Gemini 2.0 Flash Thinking Experimental what number of Rs were within the word “strawberry,” it said “two.”
Your mileage may vary.
In the wake of the publication of o1, there was an explosion of argumentation models from competing AI laboratories – not only Google. In early November, DeepSeek, a quant trader-funded AI research company, released a preview of its first reasoning model, DeepSeek-R1. That same month, Alibaba's Qwen team unveiled what it said was o1's first “open” challenger.
Bloomberg reported In October, Google reported that several teams were developing reasoning models. Below reporting Information released in November revealed that the corporate has no less than 200 researchers focused on the technology.
What opened the floodgates of the argumentative model? Well, on the one hand, the seek for novel approaches to further develop generative AI. Like my colleague Max Zeff recently reported“Brute force” techniques for scaling models now not produce the improvements they once did.
Not everyone seems to be convinced that argumentation models are the easiest way forward. For one thing, they have an inclination to be expensive as a result of the big amount of computing power required to operate them. And while they did it carried out in an excellent mood Standards It just isn’t yet clear whether argumentation models can maintain this pace of progress.
