OpenAI releases Day 2》Insane 'Reinforcement Learning Fine-tuning' new feature, enhancing AI's accuracy in professional field learning

On Day 2 of the new product launch event, OpenAI released the "Reinforcement Fine-Tuning" (RFT) feature, allowing users to enhance the inference capabilities of customized models in specific domains, thereby improving the model's accuracy. This feature can be applied in various fields including scientific research, law, finance, insurance, engineering, etc. (Previous: OpenAI releases full version o1 model and new subscription plan ChatGPT Pro, is the $200 monthly fee worth it?) (Background: OpenAI announces 12-day consecutive live streams: introducing many new features, can AI concept coins lying in ambush?) Behind ChatGPT, developer OpenAI announced a 5-day teaser, starting a 12-day new product release on the 6th at 2:00 AM. The first day featured the full version inference model o1, replacing the preview version. Currently, ChatGPT Plus, Team, and Pro users can use it, and a subscription plan allowing unlimited use of the o1 model called "ChatGPT Pro" was also introduced. Further reading: OpenAI releases full version o1 model and new subscription plan ChatGPT Pro, is the $200 monthly fee worth it? Today (7th) is Day 2 of the event, what surprises has OpenAI prepared for us? Release of new feature "Reinforcement Fine-Tuning" On the second day of the product launch, OpenAI released the new feature "Reinforcement Fine-Tuning" (RFT), allowing users to customize train the o1 model using their own datasets through reinforcement learning algorithms. Additionally, the host mentioned that Reinforcement Fine-Tuning only requires a few high-quality examples to significantly enhance the model's inference capabilities in that specific domain. It is worth noting that this feature can be applied in various fields such as scientific research, law, finance, insurance, engineering, etc. Furthermore, in the release video, Berkeley National Laboratory computational biologist Justin Reese participated in the live demonstration of the model, sharing how "Reinforcement Fine-Tuning" helps diagnose rare diseases. In the images shown by Reese, the original accuracy rate of the o1 model was 25%; the original accuracy rate of the o1 mini model was 17%, and after Reinforcement Fine-Tuning, the accuracy rate of the o1 mini model increased to 31%, achieving an 82% increase in accuracy. Although the "Reinforcement Fine-Tuning" feature significantly improves the model's accuracy and professionalism in specific domains, the host also mentioned that the feature is still in preview stage and is planned for public release next year. Additionally, OpenAI has initiated the RFT alpha program, calling for universities, research institutions, or companies to participate in testing this feature. Related reports: ChatGPT accused of "refusing to respond to David Mayer" and others, OpenAI suspected of deliberately blocking information. OpenAI announces 12-day consecutive live streams: introducing many new features, can AI concept coins be lying in ambush? OpenAI releases full version o1 model and new subscription plan ChatGPT Pro, is the $200 monthly fee worth it? "OpenAI unveils Day 2" - Revolutionary "Reinforcement Fine-Tuning" feature enhances AI's learning accuracy in professional domains. This article was first published on BlockTempo, the most influential Blockchain news media in the industry.