The 15 most important models of 2023 are as follows:
: Released by Anthropic, this is its first publicly released large language model (LLM). Anthropic, as one of OpenAI's main competitors, designed the Claude model to be as useful, honest, and harmless as possible.
: Launched by OpenAI, GPT-4 has improved upon GPT-3 and is one of the most powerful LLMs to date, surpassing human performance in multiple benchmark tests.
: The upgraded text-to-image model from Stability AI can generate images with higher resolution and superior quality.
: Developed by Meta, this AI model can isolate objects in images via zero-shot generalization.
: Meta updated its flagship large language model and released the open-source version of Llama 2. Its smaller variants (7B and 13B) provide high performance relative to their size.
: OpenAI launched an improved version of its text-to-visual model, DALL-E 3. Previously shared: 《
: Jointly developed by Google and DeepMind, this tool is used to watermark AI-generated music and images. Its watermark remains detectable even after the image has been modified. This is quite interesting, I'll check it out later.
: Launched by the French AI company Mistral, this compact 7-billion-parameter model outperforms the 13B version of Llama 2 in terms of performance, making it the top-ranked model among those of the same scale.
：Baidu, the multinational Chinese technology company, has launched Ernie 4.0, one of the highest-performing Chinese large language models to date.
：OpenAI released the upgraded large language model GPT-4 Turbo, which features a 128K context window and reduced pricing.
：OpenAI released Whisper v3, an open-source speech-to-text model, known for its improved accuracy and expanded language support.
: Anthropic launched its latest large language model, Claude 2.1, which features an industry-leading 200K context window, enhancing its ability to handle extensive content such as long literary works. Previously shared: 《
: Inflection, the startup founded by DeepMind's Mustafa Suleyman, launched its second large language model, Inflection-2, highlighting intensified competition in the LLM space.
: Google's Gemini has become a strong competitor to GPT-4, with one variant, Gemini Ultra, performing better than GPT-4 in multiple benchmark tests. Previously shared 《
: Midjourney released the latest version of its text-to-image model, enhancing user experience through more intuitive prompts and optimized image quality. Previously shared 《
In recent years, artificial intelligence systems have made significant progress relative to human baseline levels in the execution of multiple tasks. These tasks cover nine AI benchmark tests, such as image classification, basic reading comprehension, etc.
Specifically, AI has surpassed human baselines in the following areas:
first surpassed human level in this field; take the lead in the task; make breakthroughs in the field; performed better than humans.
(Competition-Level Math Problems).
We humans have not yet been completely surpassed 🐶.