Somewhat similar to the concept of the MBTI personality test, large language model technology, which strives to stay as close to human language as possible while continuously learning, has gradually developed distinct characteristics. Read on for a closer look! Inside the report on the debut of the Gemini 2.0 Flash Thinking large language model.
▲Image source for this article: Google
Think more carefully! Gemini 2.0 Flash Thinking Large Language Model arrives
Compared to the outspoken (but still more powerful than the previous Gemini 1.5 Pro) one. Gemini 2.0 FlashGoogle also announced that it will roll out even more diverse model types in the future.
And without waiting long, today they also rolled out further measures to counter the opponent’s “AI combination punch” Gemini 2.0 Flash Thinking The model is also seen as a response to competitors’ development challenges in reasoning models.
Compared to personality traits that can be simulated through prompt incantations, OpenAI released the o1 model around September, which focuses more on reasoning and has a “careful thinking” characteristic before answering questions. In short, relative to GPT-4o, which supports interrupted conversations and rapid responses, it is an LLM technology that thinks more cautiously and is better suited for specialized fields such as science, programming, and mathematics.
Now, Google is announcing that their Gemini 2.0 generation will also branch out into another type of Experimental Model, named: Gemini 2.0 Flash Thinking.
As the name suggests, the Gemini 2.0 Flash Thinking model is an experimental model that spends more time thinking. The official description states that this model is well-suited for multimodal understanding, reasoning, and code writing. The developer page also lists the following application examples:
・Reasoning on the most complex problems
・Show the model’s thinking process
Solve difficult coding and math problems.
And in the face of OpenAI, which had already rolled out a similar solution ahead of time, Google has also been extremely proactive, launching developer-facing testing in Google AI Studio under the model name “gemini-2.0-flash-thinking-exp-1219” right after Gemini 2.0 Flash Thinking was unveiled.
The Gemini 2.0 Flash Thinking model reportedly supports a context length of over 128k, with training data covering up to August 2024.
It can also be accessed through the Gemini API in Google AI Studio and Vertex AI.
Breaking news from Chatbot Arena⚡🤔@GoogleDeepMind‘s Gemini-2.0-Flash-Thinking debuts as #1 across ALL categories!
The leap from Gemini-2.0-Flash:
– Overall: #3 → #1
– Overall (Style Control): #4 → #1
– Math: #2 → #1
– Creative Writing: #2 → #1
– Hard Prompts: #1 → #1… https://t.co/lO1DiTiOOj pic.twitter.com/cq2MRMbWZ1— lmarena.ai (formerly lmsys.org) (@lmarena_ai) December 19, 2024
Interestingly, compared to when Gemini 2.0 was released—when at least the official announcement would include results from direct head-to-head matchups against their own LLMs—today’s Gemini 2.0 Flash Thinking comes with almost no official performance data. Perhaps they figured that with longer thinking time, it should be able to sweep all their own models anyway, right?
But even though the official side hasn’t released comparative data, Chatbot Arena was quick to serve up comparisons of large language models of similar caliber from Google and other brands like OpenAI (including o1!), xAI, and Anthropic. According to reports, Gemini 2.0 Flash Thinking took the top spot.
If the Gemini 2.0 Flash Thinking model really does have this kind of overwhelming performance, it seems like even OpenAI’s o1 reasoning model, which was released earlier and has received solid reviews, would be under considerable pressure, right?
Cited source:Neowin｜
Further reading:
Apple Vision Pro Taiwan Unboxing Experience, Part 2: Immersive Mode ON, Focus Mode ON
Source: KOCPC Chinese