32 mins ago
5,458
In 2024, the field of artificial intelligence large models has experienced an unprecedented rapid evolution. Just like a wonderful science and technology drama, major foreign technology giants take turns to stage breakthrough innovations. From being able to listen and speak, to being able to see and draw, to being able to make videos... AI capabilities are improving at a jaw-dropping speed.
Let’s look back on this exciting year.
Three major characteristics of the industry1. Multi-modal standardization Matching: AI from "junior college student" to "all-rounder"
Remember the earliest AI assistants? They were like students who could only do arithmetic, or could only process words. The AI in 2024 will be like participating in an "all-round training camp" and transforming into a versatile "all-round player."
Take OpenAI’s GPT-4o as an example. It can not only read text, but also understand pictures, understand voices, and understand videos.
Imagine this: you show it a photo of a street shop in Paris, and it can tell you what beauty it is, and even the production method and historical origins. This is the revolutionary change brought about by multi-modal capabilities.
In 2024, "multi-modality" that can handle multiple types of data including text, images, video and audio has become the basic standard for large models to be selected into the competition.
2. Important milestone innovation
OpenAI’s video breakthrough
In February 2024, OpenAI released its first video generation model Sora (internal beta version). This "video-shooting" AI model is a sensation in the industry. Just enter a text description and it will generate a one-minute high-definition video. Artificial intelligence is making leaps and bounds in its ability to understand and interact with real-world scenarios. Suddenly, the girl in red in the video generated by Sora displayed by OpenAI on the homepage became a "top trend".
After 10 months of hard work, OpenAI officially opened its artificial intelligence video generation model Sora to users in December.
Google's 3D world creation
In December, Google launched Genie 2, which is even more amazing. It can create an interactive 3D world from a simple picture. It's like giving AI a "magic wand" that can turn flat pictures into virtual spaces that can be explored.
Claude’s all-round upgrade
Anthropic’s Claude 3 series has made a qualitative leap in visual understanding. It can not only understand complex charts and pictures, but also conduct in-depth analysis and interpretation.
3. Faster, stronger, and more economical. A perfect balance between performance and cost
Imagine if an ordinary car can have the speed of a running car, but only needs the fuel consumption of an ordinary car. This Definitely an amazing breakthrough. A similar "technological miracle" will be achieved in the AI field in 2024. All major companies are pursuing one goal: to make AI more powerful while also making it more "energy-saving and environmentally friendly." This lays the cost foundation for the popularization of AI technology capabilities.
Let’s take a look at the specific breakthroughs:
Meta’s “Lightweight Champion”
The Llama 3.3 70B model has created the miracle of “making big things with small things” Specific performance: The speed of processing a paper is 10 times faster than before, but the cost is only one-fifth of the original. Enterprises can process more data with less budget. For example, customer service systems can serve more users at the same time.
OpenAI's "affordable version"
GPT-4o mini is like the "Spring Edition" of GPT-4o
While the cost is reduced by 97%, Still maintains good performance. One startup used the mini version to develop a chatbot, and the monthly cost dropped from $10,000 to $300.
Claude's "King of Speed"
Claude 3.5 Sonnet achieves "increasing speed without increasing price". When processing complex tasks, it has an inference speed that is 2 times higher than the previous generation model and 1. The call cost of /5 can help researchers complete a literature review in a few hours that would otherwise take days.
Competition among giants: the exciting "AI Olympics"2024 The competition in the AI field this year is as intense as the Olympic Games. Each company is like a professional player in different projects, exerting all its strength on its own "special project".
1. OpenAI: All-around Champion
Just like a decathlete in the Olympics, Open AI has shown amazing strength in many fields: Sora released in February Shocked the whole world: with just one sentence of description, you can generate a lifelike video; the Voice Engine launched in April can "clone" a speaking voice with only 15 seconds of voice samples; in December, it even started playing "razon" Innovation", releasing new products every day for 12 consecutive days.
Investment is also strong, Open AI will be launched in 2024It has received US$6.6 billion in financing and has a luxurious lineup of investors, including technology giants such as Microsoft and Nvidia.
2. Anthropic: a rising star
If OpenAI is the "veteran champion", Anthropic is a "dark horse":
Claude 3 series has won many awards in many fields Defeated GPT-4 in the test and innovatively launched the "Tool Usage" function, allowing AI to operate computers like humans. A medical institution used Claude to analyze medical records, and the accuracy rate increased by 30%. In November this year, Anthropic received another US$4 billion investment from Amazon. The two parties will build the world's largest computing cluster based on the latest chips from Amazon Cloud Technology to support the pre-training of large models, demonstrating strong strength.
3. Google: Pioneer of technological innovation
Google is like an athlete constantly challenging the limits: Gemini 1.5 broke the record of long text processing; Genie 2 realized "one picture" "Generate all things", creating an interactive 3D world from a picture, helping game developers quickly create game scenes, shortening development time from weeks to hours; the Veo 2 video generation model and enhanced version of Imagen just launched in December 3 image models are challenging OpenAI’s leadership in AI image and video generation.
4. Meta: A leader in the open source field
Meta has chosen a unique path, like a coach who publicly shares his training secrets.
The Llama series is continuously updated to benefit the open source community and make AI affordable to more people by reducing costs. Meta's open source Llama 3.2 is the first Llama model to support multi-modal input. Many small companies have developed AI applications suitable for their own needs based on Llama.
AI wins first Nobel Prize2024 Nobel Prize Of the 6 awards, awards in the fields of physics and chemistry were awarded to AI-related researchers.
American scientist John Hopfield and British-Canadian scientist Geoffrey Hinton won the Nobel Prize in Physics for their fundamental discoveries and inventions in machine learning using artificial neural networks.
David Baker of the University of Washington, Seattle, and Demis Hassabis and John Jungper of Google's DeepMind company won the Nobel Prize in Chemistry in recognition of their computational and Artificial intelligence reveals the code to the amazing structure of proteins.
One was awarded to the basic research of artificial intelligence itself, and the other was awarded to the application of artificial intelligence. These two Nobel Prizes illustrate the significance of artificial intelligenceThe huge influence of energy in the scientific field is gradually becoming more and more prominent. At the same time, artificial intelligence has accelerated from the laboratory to the real industrial field, whether it is protein biomedicine research and development, medical auxiliary diagnosis, or intelligent risk control in the financial field, and intelligent quality inspection in factories and workshops. , the capabilities of large models can be extended to all.
ConclusionAI in 2024 Development is like a wonderful science and technology movie, full of breakthroughs and innovations. From technological progress to practical applications, from giant competition to industry changes, AI is changing our world at an unprecedented speed.
AI large models continue to maintain rapid updates and iterations in terms of underlying capabilities, and the boundaries of capabilities are constantly exploring and breaking through, from text to video to 3D space, leading the development of AI in this wave. People's prospects for AGI (Artificial General Intelligence, general artificial intelligence) seem to be becoming clearer with these capability upgrades and breakthroughs.
Looking forward to 2025, the multi-modal capabilities of various AI large models are bound to continue to deepen, and personalization will also become standard. The improvement of real-time processing capabilities and the further reduction of tokens calling costs will become the capabilities of AI large models. A powerful promoter for wider application in various industries.
