News

AI Industry Newsflash

9month24

  • Step Code, support terminal programming and long mission commissioning

    13:03

    The code for Step Code for the Step Star Open AI Programming Tool. The official warehouse shows that it can read and modify project files at the terminal, run tests and process programming tasks through an interface or command line; code uses MIT licences. Step Code supports MCP, Agent Skills and plugins, providing `goal ' orders that give long-term objectives to the tool to continue, and `cron ' orders that are triggered as planned. Developer can use Step Plan account or open platform API K..

    Share to
  • Ming-Image Design Model to generate layouts and break layers

    13:03

    Ming-Image-0-1-Design and Ming-Image-0-1-Design-Layer models under ant flag. The former is a graphic model of 6B parameters, with visual design for words such as UI, information charts, posters and the like, and can generate a transparent background; the latter can break a synthetic design to multiple transparent layers as required. The official model card provides weights, reasoning and examples of two models: Design recommends 1024 or 2048 to distinguish..

    Share to
  • TIX released the latest-generation voice recognition large model

    10:51

    On 23 September, the latest generation of the mega-map of voice recognition was officially launched by TIX. It is understood that the model continues and deepens the technical capability of the Spark-Audio-1.0-Preview speech base model and achieves a significant improvement in overall speech recognition through key technologies such as enhanced self-regression synergy with LLM, a combination of Chinese and English text and acoustic enhancement, and dynamic context injection. At the same time, the reasoned cost of Spark-ASR-2.0 increased only 10% compared to the previous generation model. Sp..

    Share to
  • CEO Anthropic: The company will slow down the development of AI for safety reasons

    10:50

    In his statement to the United Nations Security Council on the subject of artificial intelligence, the CEO of Anthropic Dario Amody said that the rapid development of artificial intelligence had contributed to progress in the field of health science. Artificial intelligence can pose risks to all humanity. There is a perceived risk of illicit elements using artificial intelligence to manufacture biological weapons. If poorly managed, it is true that artificial intelligence can pose a survival risk to all humanity. For safety reasons, the company will slow down the development of AI. (FIA)

    Share to
  • Anthropic Engineer explains Claude Writing Change: Modeling is written to AI, not to people

    10:50

    Anthropic Engineer Jackson Kneen explained the decline in Claude's writing performance from Opus 4.6, which formed the so-called Claude cavity: the follow-on model optimized the focus on mathematical and code capabilities, and was heavily trained in other AI model writing techniques, which learned to write to AI rather than to humans, creating an over-intensity expression of AI that is smooth but difficult for humans to understand, rooted in an enhanced learning incentive mechanism that focuses on upgrading AI understanding effects rather than human readability. At present, Anthropi..

    Share to
  • QClaw will stop shipping on December 24, 2026, and the data can be moved to WorkBuddy

    10:49

    The local AI smart assistant, QClaw (commonly known as Lobster), announced that the business restructuring will officially cease operations on December 24, 2026 at 00:00. The registration of new users and the purchase, renewal of subscription services will cease on the date of the bulletin, and users who have introduced automatic renewal fees will automatically cease to withhold them. The service is still available to the relevant users prior to the stoppage and the credits in the account are available for use during the original period. Users with unused and valid subscription services can apply for refunds after the shutdown, with the exception of data backup and migration. Users can begin immediately until..

    Share to
  • Soybean workers are making a marginal product

    10:49

    According to media reports, the AI bean kit under the Byte flag is undergoing an internal team re-engineering, and the generic Secion team, which is responsible for the dialogue product capacity, plans to reduce by about half the staff size. According to the report, the General Security team, which originally had approximately 50 people, had been the largest in-house team of soybags, with the main responsibility for dialogue strategies, experience assessments and the like. Some of the staff will be transferred to business lines such as the commercialization of bean buns and flying books, while the rest will be abolished, and the current round will be driven by the head of the bean bag. All but product side general session..

    Share to
  • Mask predicts the future: in 20 years there may be 100 billion man-form robots

    10:48

    TESLA CEO ELLON MASK WAS INTERVIEWED ABOUT THE FINANCIAL SITUATION. WHEN IT COMES TO ROBOTS, HE LOOKS FORWARD TO A VISION OF THE FUTURE: EACH PERSON CAN HAVE AN EXCLUSIVE ROBOT, THEY CAN TAKE CARE OF OLD PARENTS, CHILDREN AND CHILDREN, AND THEY CAN TAKE ON ALL KINDS OF WORK AS A ONE-TO-ONE PRIVATE RELIGION FOR THEIR CHILDREN. IN THE FUTURE, ONE CAN MANIPULATE HUNDREDS OF, OR EVEN THOUSANDS OF, PHYSICAL ROBOTS AND DIGITAL INTELLIGENCE. “MY PROJECTIONS ARE AT LEAST 1 BILLION MAN-MADE ROBOTS IN 10 YEARS; 10 BILLION IN THE NEXT 15 YEARS; AND 100 BILLION IN THE NEXT 20 YEARS.”

    Share to
  • FORMER HUA CREATED THE YUEN MODEL CTO LEE XIAN'S YONG-GONG OPENING, AND BECAME A MULTI-BILLION-DOLLAR FINANCING COMPANY IN TWO MONTHS

    10:47

    According to the information received, Physical AI began to finance successive seed and angel ships, with a cumulative amount of hundreds of millions of yuan, with Zhong ' s assets, Zhuong ' s funds, three-flower holdings, and silver almonds capital, among others, with a company disclosure of an estimated value of $500 million. Founded in July of this year by former China, CTO Lee Chong, the current multi-mode chief scientist with the expectation that he will be a co-founder and chief scientist; and the company's primary physical infrastructure model (LPM), which addresses physical intelligence scenarios like robots. (Interface News)

    Share to
  • Google launched Gemini 3.8-series text-translation model

    10:47

    On 23 September, Google announced that the Gemini family had added two new text-translation models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, respectively. Users can expand from the original 30 original voices to an unlimited voice library. The specific functions of the model include that, once a sound is selected, both TTS models allow users to accurately control how each line of lines is presented。

    Share to
  • OpenAI: Working with Apple's ChatGPT is far less than expected

    10:46

    On 24 September Beijing time, according to the Financial Times, the latest court documents revealed how OpenAI ' s relationship with Apple had gradually broken down. On Wednesday, OpenAI stated that the company, in partnership with Apple, “showed well below expectations” after its introduction in 2024 by adding an AI function to iPhone through ChatGPT. In its legal document, OpenAI indicated that the two companies had reached an agreement in 2024 to be supported by ChatGPT “Apple Intelligence”. However, apples have failed to inspire iPhone users for this AI..

    Share to
  • Qwen-Audio-3.1 lower Qwen-Audio full voice model price

    10:46

    It is known that the Qwen-Audio-3.1 series of large voice models were officially published. This upgrade not only fully evolved the three core models of speech recognition (ASR), speech synthesis (TTS) and real-time voice interaction (Realtime), but also introduced a new audio creation model Qwen-Audio-31-TS-Next and audio interpretation model Qwen-Audio-31-ASR-Next at a heavier pound. Five brand-new voice models were launched at the same time, resulting in a complete set of audio covering "understanding-generation-interactive-creation"..

    Share to
  • Meta tested a call from the real person, Muse, causing an internal privacy dispute

    10:46

    According to information received on 23 September, according to insider posts of Meta, seen by Reuters, Meta has been testing a “manual protocol” service for its new personal AAI assistant, Muse, in which some of the calls made through the digital assistant are handled quietly by human contractors in the backstage. By means of a telephone call function, the user can instruct Muse to call the United States vendor and talk to the receiver, make an appointment for hairdressing, ask whether the door shop is available or ask different contractors for prices. These in-house posts indicate that some staff members have raised privacy concerns about the manual handling of these calls, warning that this may..

    Share to
  • OpenAI proposes to introduce third-party agencies to assess security risks in the early stages of AI model development

    10:45

    It is reported that OpenAI plans to allow third-party agencies to assess security risks at an earlier stage of the AI model development cycle, which is intended to continue efforts to address concerns surrounding potential AI hazards. OpenAI also listed the priorities it considered necessary for such an assessment to be effective, including “a strong independent mechanism, scientific rigour, robust security practices and a clear division of responsibilities”. (Interface News)

    Share to
  • Microsoft boosted the Copilot discount and launched the AI Super APP

    10:45

    According to those who know about this adjustment, Microsoft management informed the sales team this week that Microsoft authorized an increase in the discount on the Copilot artificial smart software subscription to 30% to 50% for business clients who purchased large quantities of seats and committed to paying additional fees based on the use of a specific function. According to the source, the discount would be effective from October, just as the Microsoft plan introduced a modified version of the Copilot super-application. Those who know about the program say that the new version of Copilot is as fast as this week, integrates the AI code tools, OpenClaw smart agents, etc. in a single application, and partially applies..

    Share to
  • IA price battles are all over the world: collective price reductions such as Microsoft, Amazon and Adobe keep corporate clients

    10:45

    Software companies and cloud service providers such as Amazon, Microsoft, Figma, Workday and others have been active in introducing new discounts for AI products to clients and consulting partners. Clients have been exhausted by previous changes in pricing patterns. The client had previously increased its budget for the purchase of products such as Claude Code in Anthropic and Codex in OpenAI, while cutting other expenses to cover this additional expenditure. Businesses are now becoming more cautious in selecting AI tools. The change to an AI service charge by the software vendor based on usage or number of tasks performed may push up costs and many..

    Share to
  • MASK LAMENTED CHINA'S BIG AI MODEL, SAYING THAT UNIT OUTPUT PERFORMANCE WAS ALMOST THE TOP LEVEL OF THE WORLD

    10:44

    ON 23 SEPTEMBER, TESLA CEO ELLON MASK RECEIVED AN INTERVIEW WITH ZOHANG. TURNING TO CHINA ' S AI DEVELOPMENT, HE SAID THAT CHINA ' S LARGE AI MODEL WAS, IN GENERAL, EXCELLENT. THE PERFORMANCE OF UNIT-BASED OUTPUT IS ALMOST THE HIGHEST LEVEL GLOBALLY. “I THINK THAT CHINA WILL BE ABLE TO SOLVE THE MATH MORE QUICKLY THAN MOST PEOPLE WOULD HAVE EXPECTED. A ROUGH ESTIMATE IS THAT, IN ABOUT TWO TO THREE YEARS, CHINA WILL BE ABLE TO MAKE UP FOR THE SHORTFALL BY USING LIGHT-SCRAP TECHNOLOGY AND CHIP MANUFACTURING

    Share to
  • DeepSeek published a paper signed by Liang Wen Sing to elaborate on Agent training sandboxes

    10:44

    DeepSeek published a paper signed by Leung Sing, which was first made public for the RL training and evaluation of the Agent Training Box platform DSec, V3.2 to V4.1, serving approximately 3 million sandboxes on a single-day basis, with a peak of over 380,000; the platform to unify the SDK management function, packagings, microVM and the complete virtual machine four types of backends, to decompos the mirrors into a combination-readable layer and, as required, rollout and GPU training decoupled; and the paper records Agent ' s fraudulent RPC request, modification/bin/bash, currently using AppArmor and eB..

    Share to
  • Founder of Loveable: Senior trade veterans are the largest beneficiaries of AI programming

    10:43

    Loveable CEO Anton Osika claims that it has gone through the Vibe Coding phase, that its 55% users have more than 11 years of professional experience, that 80% builders have identified themselves as non-technical; that those who understand the business can do the prototype validation and export GitHub to the engineering team to take over, and developers move to structural, safety and production responsibilities; that Loveable completed $400 million in financing in August, valued $13.3 billion, and that Anton believes that when code costs fall, creativity and selection are more critical, it is important to find users who are willing to pay for it and then get Agent to work..

    Share to
  • a16z incubation college receives $42 million to prepare the next generation of entrepreneurs

    10:43

    homeitz Andreasen Academy, a 16z hatched, was founded in San Francisco with a first tranche of $42 million, with approximately 50 people, free of tuition, no degree and credit, taking in “proof of work”; 10 companies, such as students, promoting personal projects and entering the company Co-op, Anthropic, OpenAi and NVIDIA, were founding partners, each of whom received more than $50,000; schools, for profit, planned to launch a two-year project in 2028 and close to top private universities,a16..

    Share to

Search