Notes AI employs transfer learning technology and pre-training model to support real-time adaptation of more than 200 writing styles. Its style transfer model has been trained using 5 million cross-domain text samples (e.g., academic papers, news reports, fiction books, promotional texts, etc.), and its style recognition rate is up to 91%. The median semantic similarity score of the generated text and the target style was 87/100. For example, one of the global news agencies used Notes AI to improve the productivity of the production of financial reports by 40%, reduce the rate of manual editing by editors from 35% to 12%, and ensure the density of technical words (such as “GDP growth rate” and “monetary policy”) to industry standards with an error rate below 0.5%. In the field of creative writing, after Notes AI was rolled out on an online literature platform, the word count generated by writers every day increased from 8,000 words to 15,000 words, and as per the platform’s data, the user retention rate in AI-assisted chapter writing (72%) is 14 points higher than that of fully manual writing (58%).
At a technical level, Notes AI uses a hierarchical attention mechanism to analyze the stylistic features of the user input (e.g., sentence difficulty, word distribution, emotional strength), and its parameter tuning module offers real-time adjustable adjustment of generation methods. For example, on the example of legal documents, Notes AI analyzed three frequency of keyword occurrence aspects of clauses of a contract (e.g., frequency of occurrence of “liability” and “force majeure” ≥5 times/thousand words) and logical structure, and the rate of compliance-generated text pass improved from 68% to 93%, and a law firm saved 70% of drafting time for a contract, and the error ratio decreased to 0.8%. In addition, with the integration of Notes AI’s sentiment analysis model (89% accuracy) and buzzword database (100,000 hot words) into short social media postings (e.g., Twitter and Weibo), the interaction rate of brand marketing copy increased by 22%. After a FMCG firm’s usage, the peak retweet times reached 120,000 times, 3 times higher than that of manual composition.
According to the 2023 Natural Language Generation Technology White Paper, Notes AI excelled in hybrid multi-style tasks, such as mimicking both the hybrid needs of a technical report (with data graph description error +/−1.2%) and satire writing (with emotional polarity alignment 85%), with contextual consistency score of 79, bettering the industry standard (65). In teaching, when an online tutoring platform for writing adopted Notes AI, the two models of “academic rigor” and “narrative vitality” in the student essays were embraced with 78% and 82% adoption rates, respectively, and the correction efficiency of teachers was increased by 50%. However, the current system’s adaptation to extremely low frequency styles (such as 19th century classical composition) is still biased, and word generation in the text has an error rate of up to 15%, which is incremental training optimization based.
According to market feedback, Notes AI’s own style adaptation ability (which allows companies to upload more than 500 reference texts) can reduce the model adaptation time from 30 days to 72 hours, and a cross-border e-commerce company can take advantage of this to facilitate a localized style conversion of product description copy, which increased the click rate in Southeast Asia market by 18% and reduced the return rate by 6%. Technology firm Gartner anticipates that AI software for multi-style adaptation will make up 75% of enterprise content creation use cases by 2026, and Notes AI has become among the leading solutions in the market due to its high accuracy (90% recovery of style features) and low cost ($0.003 per generation).