1. baichuan Intelligence launched 7 billion Chinese and English pre-training large model Baichuan 7B, providing 1.2 trillion token, 10,000 level ultra-long dynamic window and deep optimization training process, which is expected to be applied in many fields such as text generation, text classification question-and-answer system, and machine translation.
  2. Baidu Wenxin Yige and Jingdong, through AI technology to achieve the rapid generation of offline advertising creativity, saving time and cost, improve the freshness
  3. Qualcomm demonstrated the mobile phone 15 seconds run 100 million large model: The team specifically based on mobile AI hardware and software technology, the open source model was optimized, according to Qualcomm, in this wave of generative AI wave, compared with the cloud computing power processing AI model, the end of the computing power can also be used
  4. Sensetime released the Go robot, which can reach the highest level of professional nine-dan with more than 20,000 exercises, to meet the needs of chess from enlightenment to amateur five-dan.
  5. Tao Zexuan himself used GPT-4 full blood: Before the release of GPT-4 obtained access from Microsoft, Microsoft invited him to experience after writing an article, arguing that human expectations of information technology all need to be calibrated
  6. Microsoft introduced the innovation of integrating ChatGPt-like Copilot into the ERP product matrix to help enterprises achieve cost reduction and efficiency, accelerate project development, and improve productivity and customer experience.
  7. AMD released the artificial intelligence chip M1300X, A rival to Nvidia’s GraceHopper series, with a staggering 153 billion transistors and 192GB of HBM3, opening the A model to larger scale computing.
  8. FINE-GRAINED human feedback reinforcement learning with RLHF: fine-tuning works better than ChatGPT, saving language model nonsense
  9. GPT-4 passed the MIT undergraduate math exam with full marks. The original version of GPT-4 was ready to use out of the box, without any tuning at all, and got a 90% score in this MIT exam
  10. Meta CEO Mark. Zuckerberg is considering commercializing a new version of LLaMA, an open-source large-scale language model, to allow other companies to develop AI software using LLaMA for free, allowing developers to use their products for commercial use.