AI trend 2025 [A]
Here are the contents translated into English from the
2025 will be a year of significant development for artificial intelligence. According to IBM’s latest report, AI knowledge is likely to become a necessary prerequisite for many jobs soon. Regardless of industry, you’ll need some knowledge of AI. The fourth trend is that markets will find new ways to build more powerful AI.
Another certainty in 2025 is that OpenAI will officially release its reasoning model o3. o3 can already solve research-level math problems with the same accuracy as a competition-level programmer handling coding tasks.
ChatGPT 4’s IQ is 115, equivalent to the average intelligence of a university student. O1 pro has an IQ of 139, making it a rare talent. Meanwhile, o3’s IQ is 157, which means only one in 13,333 people are as intelligent. OpenAI plans to release a mini version of o3 in late January 2025, followed by the full version.
The reasoning model will significantly accelerate scientific research progress. Moreover, it’s worth noting that OpenAI developed o1 to o3 within just a few months.
Perhaps in 2025, mathematicians will use o3 or o4 to prove a famous conjecture, which will have a profound impact on human society.
We may see the emergence of super-large language models. Currently, markets predict that the next-generation model will be many times larger than its predecessors, potentially exceeding 50 quintillion parameters.
If we can expect super-large language models in 2025, we might also see small language models specifically designed for certain tasks. These models would have only tens of billions of parameters and could run on a laptop or even a smartphone without requiring massive GPU data centers.
This means that these models rely on less data and are more energy-efficient. They can be deployed on personal devices.
This is the fifth trend: we’re getting closer to achieving general-purpose artificial intelligence (AGI). Let’s start with the basics. AGI, short for Artificial General Intelligence, refers to a machine that can think and learn like humans. It can perform various tasks, such as helping you solve problems, learning new knowledge, and communicating with people.
Moreover, it doesn’t require special instructions or settings. AGI can adapt and understand complex situations like humans.
In fact, some AI models and technologies already have the ability to do things at an AGI level. For instance, chatbots like ChatGPT can remember every conversation you’ve had with them and have a growing memory window.
This means we’re soon entering the era of customer service chatbots. AI teachers will be able to track students’ progress throughout their studies, while AI health consultants will analyze your medical reports and create personalized diet plans.
Research has shown that AI diagnosis is already at a level where it surpasses human doctors and ChatGPT combined. In other words, humans can’t keep up with AI’s capabilities.
OpenAI’s CEO Sam Altman recently wrote in his blog that they’re confident about building AGI now. In 2025, we may see the first batch of AI intelligent entities join the workforce. Even he mentioned that OpenAI is shifting its focus towards creating truly super-intelligent machines.
The above are the five most important trends I’ve identified based on my research for artificial intelligence in 2025. My predictions are the minimum expectations; there will certainly be other surprises and groundbreaking events in 2025, perhaps even some teams working secretly to revolutionize our understanding of AI.
I’ll also keep you updated with the latest news from the AI market and share tips on improving work efficiency using AI software.
Translation
2025年是人工智能发展的一年。根据IBM最新的一份报告显示，人工智能知识可能很快就会成为很多工作的必备先决条件。无论你在哪个行业，都需要懂些人工智能。第四个趋势是市场将找到新的方法来构建更强大的人工智能。
2025年另外一个确定性就是OpenAI将会正式推出推理模型o3。 o3已经可以解决研究级的数学问题，同样以竞赛级程序员的准确性处理编码任务。
ChatGPT 4 o 的智商是115，相当于一般大学生的平均智商。 O1 pro的智商是139，已经是百里挑一。 而o3的智商则是157，也就相当于每13,333人中才有一个这么聪明的，也可以称为万里挑一。
OpenAI计划在2025年1月底先推出o3 mini版本，后续还会推出o3的完整版。推理模型将大大加速科学研究的进度。同时，你别忘了OpenAI从o1到o3才用了短短几个月时间。
也许在2025年，会有数学家借助o3或者o4证明一个著名的猜想，这将对人类社会产生重大震撼。
所以，我们可能会看到一些超大语言模型的诞生。现在市场预计下一代模型将会比之前的版本大很多倍，也许会超过50万亿个参数。
如果说2025年有可能出现超大语言模型，那我们还可能看到针对特定任务进行微调的小模型，即small language model。这类模型的大小只有几十亿个参数，而且它们不需要装载大量GPU的大型数据中心来运行。
这意味着这些模型依赖更少的数据，更加节能，它们可以在你的笔记本电脑甚至是手机上运行。
这也是第五个趋势，即我们将越来越接近生成式人工智能AGI。首先，让我们普及一下AGI，全称是Artificial General Intelligence，也就是通用人工智能。这是一种能像人类一样思考和学习的智能机器，它当然可以做很多不同的事情，比如帮助你解决问题、学习新的知识与人交流。
并且，它不需要你给它特别的指令或者设置。AGI能像人类一样，全程适应和理解各种各样的情况。
事实上，现在有的AI模型和技术已经能做一些AGI级别的事情了。举个简单的例子，像ChatGPT这样的聊天机器人已经可以回忆起与我们进行过的每一次对话，它的记忆窗口也会变得越来越长。
这意味着我们很快就会进入客服聊天机器人的时代，这就意味着AI老师能够全程跟踪学生的成长进度，AI还能做个人的医疗顾问查看你的体检报告，为你制定健康饮食计划等等。
我听说，有研究发现AI诊断的水平目前已经高到，如果让人类医生和ChatGPT一同诊断，结果会不如让ChatGPT自己诊断。换句话说，人已经帮不上什么忙了，也许我们应该全听AI的。
OpenAI的首席执行官Sam Altman在最新的一篇博客中也表示，他们现在有信心知道如何构建AGI。在2025年，我们可能会看到首批AI智能体加入劳动力市场。甚至他补充到，OpenAI开始将他们的目标转向真正意义上的超级智能。
以上就是我根据调研结果给你解读的2025年人工智能最重要的五个趋势。当然，我说的这些是最起码的期待，2025年肯定会有一些其它的惊喜，也许是颠覆我们认知级别的大事件，也许某个团队正在默默地憋大招，那就让我们拭目以待。
同样，我也会在第一时间，为你带来AI市场的最新资讯解读，以及我找到的提高工作效率的AI软件。手把手教你如何用，提高工作效率，告别牛马生活，我们下期影片见拜拜！
Reference:
https://www.youtube.com/watch?v=eDAF8ckt1J8