aifollow.news

All AI updates

Primary

Oct 29 Today Oct 29 Thursday

Thursday · 20 items
IT之家 科技新闻 Machine translation

Samsung launches W27 foldable phone starting at 18,999 yuan

The W27 is based on the Galaxy Z Fold8 Ultra and comes in 16GB+512GB and 16GB+1TB configurations priced at 18,999 yuan and 21,999 yuan, respectively. After activating the Tiantong satellite service, users receive one year of direct satellite connectivity, 300 minutes of satellite calls and 100 satellite texts. The phone also supports AI features including Circle to Search and live translation.

IT之家 科技新闻 Machine translation

South Korea’s Education Ministry proposes a three-year exam ban for AI glasses cheating

South Korea’s Education Ministry has announced a plan to curb cheating on the university entrance exam. It proposes extending the exam ban for cheating with smart glasses from one year to three years and plans to pursue criminal charges against those who use them to cheat. Each test site will assign a specialist to identify smart glasses starting this year; signal-detection equipment remains in development, with deployment next year subject to evaluation.

IT之家 科技新闻 Machine translation

Momenta previews R7 Ultra world model, claiming fivefold parameter scale

Momenta announced at an XHEART event that it will release the R7 Ultra world model, claiming five times the previous parameter scale and a fivefold improvement in safety performance. The model is intended to run in vehicles using the X7 Ultra chip. Momenta says it can handle complex driving scenarios including U-turns, passing encounters and parking; these capabilities remain the company's claims.

英伟达中文博客 Machine translation

European financial firms showcase open model and accelerated computing applications

Revolut says its PRAGMA foundation model is serving an initial 80 million customers in production. Starling Bank has developed an intelligence layer using its transaction data, Lloyds Banking Group has built a transaction foundation model, and SBS has launched a platform for running models within banks' own environments. bunq says its open-weight assistant costs 60% as much per answer as the closed models it compares against, while NatWest, Modulr and Santander are developing speech or finance-specific models. LSEG reports up to 10 times faster transaction-data analysis, and RavenPack has built a research dataset pipeline. HSBC reports five times faster data processing; in an offline test, Stripe reports a 42-fold faster data pipeline at 92% lower cost.

IT之家 科技新闻 Machine translation

Anthropic launches Claude reporting tool for enterprises to track usage and costs

Claude Enterprise subscribers can now enable the reporting tool in organization settings to review usage, output types and costs by team, product and a time range of up to 28 days. Organizations can generate up to 10 reports per month for free. The tool reads team conversations and shows session details; Anthropic says it should not be used to assess individual performance or make personnel decisions.

IT之家 科技新闻 Machine translation

Zhiyuan plans to add perception models to robots, Peng Zhihui says

Zhiyuan co-founder Peng Zhihui said many dancing robots still imitate motions without perception, and the company plans to introduce perception models. He said its previously released integrated perception and control model can transfer the ability to handle complex terrain across robot forms, including humanoid and quadruped robots.

MarkTechPost AI 研究报道 Machine translation

Block releases research preview of Dessa, a speech-only full-duplex model

Dessa can listen while it speaks. Block says it trained the model only on human conversation audio, without text or pretrained models. The report says it runs locally on a MacBook Pro M4 Max or better, but access is currently by waitlist; no benchmark scores, weights or API have been published.

IT之家 科技新闻 Same story Machine translation

NVIDIA releases Nemotron 3.5 Transcribe weights and Granary v2 speech data pipeline

Nemotron 3.5 Transcribe is available for free on Hugging Face and ModelScope. It can process recordings up to one hour long, transcribe 40 languages and distinguish up to eight speakers. The Granary v2 pipeline, released alongside it, covers 463,062 hours of audio across 48 languages and can turn audio into training data for transcription, translation and other tasks.

3 more sources reported this
  • MarkTechPost AI 研究报道
  • TechCrunch AI
  • Hugging Face 官方博客
Same event; featured as “NVIDIA 发布 Nemotron 3.5 Transcribe 多语种转录模型与 Granary v2 数据集”
MarkTechPost AI 研究报道 Machine translation

AssemblyAI releases TTS-1 API for live voice agents at $15 per million characters

AssemblyAI-TTS-1 is generally available as a hosted API for voice agents on live calls. It costs $15 per million characters, with no concurrency tier to buy. It supports six languages and instant voice cloning, but runs only in AssemblyAI's US or EU cloud and does not yet support SSML, speed control, or custom pronunciation dictionaries.

IT之家 科技新闻 Machine translation

South Korean court rejects 89 Tesla owners’ FSD refund claims

Eighty-nine Tesla owners in South Korea sought refunds because they could not use the full FSD package, but the Seoul Central District Court rejected their claims. The court found that Tesla had delivered some features and was required to provide FSD functions only where legally permitted. Regulatory and software deployment restrictions still prevent some local owners from using features such as city navigation and automatic parking.

IT之家 科技新闻 Machine translation

STMicroelectronics expects fourth-quarter revenue growth; Texas Instruments plans to expand capacity

STMicroelectronics expects quarterly revenue of $4.15 billion at the midpoint, above analysts’ $4.11 billion average forecast. Its projected gross margin of 39% also exceeds analysts’ 38.2% estimate. The company is targeting demand for power-management chips in AI data centers. Texas Instruments issued an upbeat revenue outlook and plans to expand capacity, while NXP Semiconductors issued a revenue outlook that fell short of expectations.

IT之家 科技新闻 Machine translation

Instagram chief Mosseri expects public anger over AI to grow

Mosseri expects public dissatisfaction and anger over AI to intensify in the coming years, affecting people beyond the creative industry. He confirmed that Instagram will not treat creators differently for using AI, but questioned whether it can rank AI content as effectively as non-AI content. Instagram has begun requiring AI accounts to identify themselves.

IT之家 科技新闻 Machine translation

Claude Code 2.1.311 adds setting to bypass further dangerous-operation checks

Anthropic added an environment variable in Claude Code 2.1.311 that, when enabled by a user, can disable some dangerous-operation blocks retained by the existing permission-skip option, such as the block on recursively force-deleting a user's home directory, according to the report. The setting is not enabled by default, and other restrictions imposed by organization administrators may still apply.

IT之家 科技新闻 Machine translation

SpaceXAI plans to buy homes near data center power plant, requiring sellers to waive noise lawsuit claims

SpaceXAI plans to buy homes near a gas turbine plant that powers its data center, according to a report. The plan covers 418 homes, and at least 18 owners have agreed to sell, though the company has not confirmed those figures. Purchase letters require sellers to waive claims in a noise class action but give no specific offer. SpaceXAI denies the allegations.

Latent Space AI Same story Machine translation

Google DeepMind launches Gemini 4 Carbon for trusted cyber defenders first

Google DeepMind launched Gemini 4 Carbon, with initial access for trusted cyber defenders through the Fairwind Program and wider developer access promised as soon as possible. Artificial Analysis ranks it first on Intelligence Index v5 with a score of 59, though it scores 59 on the firm's Coding Agent Index, below Claude Code with Opus 5.5 at 62.

6 more sources reported this
  • 量子位 AI
  • ginobefun (@hongming731)
  • IT之家 科技新闻
  • The Verge AI
  • MarkTechPost AI 研究报道
  • Rohan Paul (@rohanpaul_ai)
Same event; featured as “Google 发布 Gemini 4 Carbon，先向部分网络防御人员开放”
IT之家 科技新闻 Machine translation

Tencent and Visa sign memorandum to explore cross-border and agent payments

Tencent and Visa have signed a memorandum of understanding and plan to improve cross-border payments for inbound and outbound travelers through TenPay Global and Visa Pay, subject to regulatory requirements. They will also explore AI agents helping eligible users choose and buy goods or services, while Tencent Cloud and Visa will explore cooperation in enterprise AI and other areas.

IT之家 科技新闻 Machine translation

XREAL AURA goes on sale overseas, starting at $1,099 for a limited time

XREAL AURA is now on sale overseas, with initial shipments expected in mid-November. The glasses use an external computing module and run Android XR. They launch with more than 150 XR experiences, and users can access existing Android apps through Google Play.

IT之家 科技新闻 Machine translation

Tencent says WorkBuddy Enterprise covers 50% of China’s top 500 companies and plans to open its product architecture

Tencent executive Tang Daosheng said WorkBuddy Enterprise has reached more than 50 industries and 50% of China’s top 500 companies since its March launch. Tencent said it plans to open WorkBuddy’s product architecture to company-built and third-party agents, including connections to third-party IM systems; the opening has been announced but is not yet described as available.

IT之家 科技新闻 Machine translation

South Korea plans at least 5 trillion won to expand physical AI adoption among small manufacturers

South Korea plans to invest at least 5 trillion won by 2030 under a strategy combining digitalization, AI and robotics, aiming to raise AI adoption among small and medium-sized manufacturers from 1% to 10%. It also aims to increase the number of AI-adopting firms from 600 to 6,000, bring manufacturing robots to more than 1,000 firms, and tailor support to companies’ existing data capabilities.

Load more