Babbitt丨Daily must-read: iFLYTEK Spark Cognitive Model V3.0 was officially released; Apple plans to integrate AI elements for its services and platforms

Abstract: iFLYTEK Spark Cognitive Large Model V3.0 is officially released. It is understood that the current Spark model has surpassed ChatGPT as a whole. Apple introduced automatic error correction based on the "Transformer" model in the iOS 17 system, indicating that Apple began to join the generative AI battle.

Image source: Generated by Unbounded AI

Hot News:

iFLYTEK: Officially released Spark Cognitive Model V3.0, surpassing ChatGPT

Babbitat News, on October 24, the 2023 iFLYTEK Global 1024 Developer Festival opened in Hefei. iFLYTEK announced that iFLYTEK Spark Cognitive Model V3.0 was officially released. It is understood that the current Spark model has surpassed ChatGPT as a whole.

Apple expands hiring efforts to integrate AI elements into its services and platforms

According to the IT House news on October 24, Apple introduced automatic error correction based on the "Transformer" model in the iOS 17 system, indicating that Apple began to join the generative AI battle. According to Apple's official recruitment list, the company has increased the recruitment of AI engineers, planning to integrate AI elements for many of its services and platforms. For example, a job that is planned to be recruited for the App Store reads in its introduction: "An AI-based generative developer experience platform is being developed for internal use and to assist our app development team." Another effort by Apple's retail division mentions developing "conversational AI platforms (voice and chat)" to engage with customers. Apple's job listings also mention tasks such as building text generation technology, such as "long text generation, summaries, Q&A."

Moody's and Google Cloud to collaborate to develop generative AI applications for financiers

According to PRNewswire, on October 24, Moody's and Google Cloud announced a new strategic partnership to explore combining Moody's expertise in financial analytics with Google Cloud's advanced generative artificial intelligence technology to help Moody's customers and employees leverage the new Large Language Model (LLM) to gain new financial insights and summarize financial data faster.

Specifically, the partnership is premised on three core objectives, including building an LLM to accelerate financial analytics; Allows access to Moody's data in BigQuery for financial insights; Enhance enterprise search for financial data.

iFLYTEK released the Spark Medical model, which comprehensively surpasses the GPT-4 in medical treatment

According to the Securities Times, on October 24, iFLYTEK released the iFLYTEK Spark Medical Large Model at the 2023 Global 1024 Developer Festival, and launched the "iFLYTEK Xiaoyi" App and Mini Program with the blessing of the medical large model. It is reported that the large model of iFLYTEK Xinghuo Medical has passed the evaluation of standards and specifications organized by the Academy of Information and Communications Technology and the National Health Commission. Liu Qingfeng, Chairman of iFLYTEK, said that the Spark Medical model has comprehensively surpassed GPT-4 in terms of medical mass knowledge Q&A, medical complex language understanding, medical professional text generation, and question answering rate recommended for medical diagnosis and treatment.

iFLYTEK and Huawei jointly build the computing power base "Feixing No. 1", iFLYTEK Spark 4.0 will be fully benchmarked against GPT-4

According to IT House reported on October 24, Liu Qingfeng, chairman of iFLYTEK, announced at the 2023 Global 1024 Developer Festival that the company has joined hands with Huawei to build the "Feixing No. 1" platform, an independent and controllable large model computing power base based on the Ascend ecosystem, on this basis, iFLYTEK Spark Large Model will start a larger scale training against GPT-4. He also said that in the first half of 2024, the release of iFLYTEK Spark 4.0 will be fully benchmarked against GPT-4.

At the same time, Liu Qingfeng said, "Compared with GPT4 with trillion parameters, there is still a gap between all large models and it, and we must face the gap with GPT4, and need to strengthen it from the aspects of complex knowledge reasoning and small sample rapid learning." The battle of the big model is definitely a tough battle, by no means can be done by bragging, only the spirit of seeking truth from facts can truly achieve transcendence. ”

iFLYTEK responds to inappropriate content on learning machine: has put the large model content review mechanism on the learning machine to strive to be more stringent

According to the interface news on October 24, recently, some parents claimed that some parents found some content that defamed great people and distorted history in their children's iFLYTEK learning machine that violated mainstream values. In response, Liu Qingfeng, chairman of iFLYTEK, responded that the content has been removed as soon as it was discovered, and the partners responsible for ensuring content security have been punished, and the content moderation mechanism used for large models has been placed on the learning machine, striving to make the content review of the learning machine more stringent.

Ctrip: Further improve the Q&A model and integrate generative AI into various products

According to Wall Street News on October 24, Ctrip's Global Partner Summit was held in Singapore today, and Liang Jianzhang, co-founder and chairman of the board of directors of Ctrip Group, announced the strategic direction of the triple innovation of generative AI, content list, and ESG low-carbon hotel standards.

According to reports, in terms of AI, Ctrip has further improved Ctrip's Q&A model, integrating generative AI into various products, and using an interactive interface based on natural language to improve user experience and help merchants reduce costs and increase efficiency.

OCBC Singapore launches generative AI chatbot for global workforce

According to the Webmaster's House on October 24, Singapore's Oversea-Chinese Overseas Chinese Bank (OCBC) became the first bank in Singapore to introduce generative AI chatbots to its global workforce. According to official announcements, OCBC will make generative AI chatbots available to its 30,000 employees worldwide in November 2023.

The chatbot was developed in collaboration with Microsoft's Azure OpenAI to assist employees with writing, research, and ideation. OCBC ChatGPT is powered by ChatGPT's "large language model," which generates answers and results based on textual information available on the web, while remaining in a secure and controlled environment. During a six-month trial from April to September, OCBC's 1,000 employees used the bank's GPT capabilities in areas such as investment research reports, content translation, and draft customer responses. Including the time required to ensure factual accuracy, tasks are completed 50% faster than before.

Foreign media: Samsung Galaxy S24 series will be fully invested in artificial intelligence, planning to become the smartest AI mobile phone

As reported by Sammobile on October 24, Samsung plans to fully invest in artificial intelligence technology on its next-generation flagship phone, the Galaxy S24 series. Samsung wants the Galaxy S24, Galaxy S24+, and Galaxy S24 Ultra to be the smartest AI phones ever, surpassing even the features Google offers on the Pixel series.

It is reported that Samsung officially announced the Exynos 2400 processor at the LSI Technology Day event held in San Jose, California on October 5 this year, saying that the CPU performance is 70% faster than the Exynos 2200, and the AI processing capacity is 14.7 times faster.

Kuo Ming-chi estimates Apple will spend $4.75 billion on 20,000 AI servers next year

According to the "Science and Technology Innovation Board Daily" reported on October 24, Guo Mingxi said that it is estimated that 2,000-3,000 units will be purchased in 2023, accounting for 1.3% of global AI server shipments. In 2024, 18,000 to 20,000 AI servers will be purchased, accounting for 5%. Kuo Mingxi believes that the AI servers purchased by Apple are mainly the most common NVIDIA HGX H100 8-GPU for training and inference generative AI, and the purchased chip in the fourth quarter of next year will join the B100. The H100 is calculated at a price of $250,000, and Apple's purchase cost on AI servers in 2023 is estimated to be $620 million, and it is estimated to reach $4.75 billion next year.

Nearly half of ByteDance's PICO employees left, and many executives left and transferred

According to The Paper on October 24, ByteDance is reconsidering the strategic positioning of ByteDance's VR (virtual reality) business PICO. It was learned from a number of people familiar with the matter that many of PICO's core business leaders have recently left and transferred jobs, and the number of employees in PICO's previous peak period exceeded 2,000, of which about 2/3 of the employees in the marketing department left, and the turnover rate of the technical department was slightly lower, and the overall number of employees leaving at present was close to half of the peak period. The relevant person in charge of PICO replied that the company will maintain the normal operation of products and continue to strengthen investment in product technology.

Previously, it was reported that "VR giant PICO will be gradually shut down, and ByteDance will abandon the meta-universe", but the relevant person in charge of ByteDance responded to the rumor: this news is not true, PICO is operating normally, and the company will invest in XR (extended reality) business for a long time.

Investment and Financing:

UK AI cybersecurity startup Harmonic completes $7 million funding round

As reported by Webmaster House on October 24, British AI startup Harmonic Security recently raised $7 million to build technology to secure enterprise generative AI deployments. Harmonic will develop software to address the problem of unregulated AI applications collecting large-scale company data.

The start-up funding was led by Ten Eleven Ventures, and Harmonic founder Alastair Paterson (who led Digital Shadows' $160 million acquisition by ReliaQuest/KKR) said their technology promise provides enterprises with a complete picture of generative AI adoption, providing risk assessment for all AI applications and identifying potential compliance, Security or privacy concerns.

AIGC Recommendation:

New Big Model Claims to Outperform GPT-4 at Every Turn, We've Compiled These Review Tools

Since the advent of ChatGPT, there has been an "arms race" for big models around the world. According to reports, from January to July this year, a total of 64 large models were released in China. As of July 2023, a total of 130 large models have been released in China. "Hundred model war" is not enough to describe today's scorching "war situation", so which big model is better? This is inseparable from the evaluation of large models.

** "Even ChatGPT understands "sycophancy"! OpenAI's Strongest Competition: It's All the Mistakes of "Human Preference"**

Have you ever thought that ChatGPT-generated answers are influenced by the user's personal preference to respond to something "sycophancy" enough to be a neutral or truthful message? In fact, this phenomenon is present in most AI models, including ChatGPT, and the culprit may be "reinforcement learning based on human feedback (RLHF)".