Upgrade and launch of APP and mini program iFLYTEK said it accelerated benchmarking against ChatGPT

Chen Jialan, a reporter from this newspaper, reported from Guangzhou

After a month, iFLYTEK upgraded its own large-model products.

On June 9, the reporter of "China Business News" learned from iFLYTEK that iFLYTEK Cognitive Model V1.5 was launched, and Liu Cong, Dean of iFLYTEK Research Institute, demonstrated in real time the capabilities of iFLYTEK Spark Cognitive Model V1.5 in open-ended knowledge quizzes, logical reasoning, mathematics and multi-round dialogue, as well as the improvement of capabilities in learning, office, medical and industrial fields. At the same time, the iFLYTEK Spark APP is online, however, at present, only the Android version can be downloaded.

Previously, on May 6, iFLYTEK released the Spark cognitive big model, which demonstrated to the outside world in real time, including text generation, language understanding, knowledge question and answer, logical reasoning, mathematical ability, programming ability, multi-modal and other capabilities.

"Today we are releasing the latest results of the Xinghuo model again as scheduled, because all our technology platforms are completely independent research and development, and we are very confident that the technology at every stage is controllable." Liu Qingfeng, Chairman of iFLYTEK, said that iFLYTEK should not only pay tribute to and catch up with OpenAI, but also do independent innovation at the source, and has explored more potential paths and cutting-edge cross-research opportunities for general artificial intelligence in brain-like intelligence, neural network large models, and game intelligence.

Three aspects of capability upgrade

The reporter learned that iFLYTEK Xinghuo cognitive big model V1.5 has been upgraded in three aspects: open-ended knowledge question and answer, logical reasoning, mathematics and multi-round dialogue.

Liu Cong asked the newly upgraded Xinghuo Cognitive Model V1.5 "What are the essay questions in the first volume of the 2023 National College Entrance Examination, and analyze the meaning it conveys". "The college entrance examination has just ended, and the child is about to start a new beginning, please write a warm letter to your child." After a while, the big model answered according to his own understanding.

Specifically, in terms of open-ended Q&A, due to the inherent operating mechanism of the large model, the update of new knowledge is a difficult problem, and the timeliness of the knowledge contained in the provided answers is often insufficient. Wu Xiaoru, President of iFLYTEK, said that by combining the language understanding ability and summary expression ability of the large model with the search plug-in, the Spark model has effectively solved the industry problems such as the difficulty of updating new knowledge and the easy "Zhang Guan Li Dai" of fact-based Q&A, and achieved a 24% improvement in knowledge Q&A ability.

"The real power of the big model is not that the generated content is exactly the same as direct search, only with natural language understanding as the core, combined with plug-in capabilities such as class search, can more complex problems be solved." Wu Xiaoru further said that language understanding ability is the most important ability for large models to become human assistants in the future, and the breakthrough of this open question and answer reflects iFLYTEK's leading language understanding ability.

With the blessing of long-chain thinking reasoning and mathematical logical reasoning ability, iFLYTEK Spark Model V1.5 has been greatly upgraded in logical reasoning and mathematical ability.

"Farmers need to cross the river with wolves, sheep and vegetables, only farmers can row, and the boat is relatively small, farmers can only bring one thing across the river at a time. If the farmer is not there, the sheep will steal the vegetables and the wolves will eat the sheep. Please devise a way for farmers to safely bring everything across the river. Faced with this complex mathematical problem, the iFLYTEK Spark model quickly gave the answer.

In addition, for polynomial arithmetic problems that are difficult to input by voice or typing, iFLYTEK Spark model has supported photo recognition to solve the problem. At the scene, Liu Cong took out a high school mathematics exam paper, faced with a complex polynomial question, Xinghuo APP selected the specified question through the photo frame, and quickly gave the correct answer on the basis of OCR recognition.

In addition, the newly upgraded version of iFLYTEK Xinghuo Cognitive Large Model V1.5 has also further improved its ability in multi-round dialogue. "I'm a graduating college student and I want to interview for a position as a product manager." The iFLYTEK Spark cognitive model immediately turned into an interviewer on the spot and had a continuous dialogue with the questioner.

Launch APP and accelerate benchmarking against ChatGPT

On May 19 this year, ChatGPT landed on the iPhone in the form of an app, which suddenly exploded the market, and the application intelligence provider data.aidata.ai According to the report, OpenAl's iOS version of ChatGPT has been downloaded more than 500,000 times in less than a week.

Not long after Baidu's Wenxin Yiyan was released, a pirated version of the "Wenxin Yiyan" APP appeared on the Apple App Store.

It can be seen that the large model landing on the mobile terminal is undoubtedly more convenient and conducive to reaching more users. This time, Spark Model launched the Spark APP and Mini Program, of which the APP supports more than 200 assistants.

In addition, the upgraded iFLYTEK Spark model has also improved its capabilities in the fields of learning, office, medical and industrial fields.

For example, the AI speaking assistant has been upgraded to the Spark Language Companion APP, and with the blessing of the large model, the Spark Language Companion APP can conduct open-ended dialogue, situational communication, sparring like a speaking teacher, and perform real-time oral error correction. Users can also use video conversations to communicate face-to-face with virtual human teachers and practice speaking in an immersive way.

For example, in the face of the current situation of 250 million patients discharged from the mainland every year, only 1.88 million doctors above the intermediate level, and the vast majority of patients are "discharged from the hospital", Wu Xiaoru pointed out that based on the iFLYTEK Spark model, it will improve the ability of post-diagnosis rehabilitation management, and through the analysis of cases, it will help doctors quickly generate rehabilitation plans and guide patients to take medication.

Liu Qingfeng revealed that there will be a new version on August 15, and the code capability will be greatly improved, and the multimodal interaction will be upgraded; On October 24, the general model of the Spark cognitive big model will be directly benchmarked against ChatGPT. In the future, iFLYTEK will also have GPT4 benchmarking products, which will meet with the outside world at the right opportunity.

"The reason why we can set such a clear goal is because we not only have a good technology accumulation, a well-established team, but also each key module is completely independent research and development, and the software and hardware platforms are all run on domestic reliable platforms." Liu Qingfeng said.

"We are not only paying tribute to and catching up with OpenAI, but also doing independent innovation at the source." It is reported that the National Key Laboratory of Cognitive Intelligence undertaken by iFLYTEK has been deployed in many fields such as brain-like intelligence, neural network large models, and game intelligence, exploring more potential paths and cutting-edge cross-research opportunities. "The new era of general artificial intelligence will be a great historical process, which will not be achieved overnight, so we must have both short-term ambitions and long-term perseverance." Liu Qingfeng said.

CSC believes that the domestic boom in the R&D and application of large models continues to rise, and the development of large models is accelerating in an all-round way, but the current global large model industry is still in the early stage of exploration, and it is necessary to cooperate with downstream scene enterprises to establish a business model of large models.

(Editor: Wu Qing, Proofreader: Yan Yuxia)