01News
ByteShape's full ShapeLearn release of Qwen 3.8 27B: five GGUF models on the measured quality-speed frontier across RTX 3090, 4080, 4090, 5060 Ti, 5090 and RTX Pro 6000. GPU-5 is the default wherever it fits, at 99.63% of BF16. GPU-4 stays very competitive at 98.72% in a much smaller file. Plus why ShapeLearn-Lite beat its KLD ranking, and MTP versus DFlash2 speculative decoding.
Source links can be verified on the original siteRead original ↗02News
Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Tesla CEO Elon Musk have successively called for slowing down the development pace of the most advanced artificial intelligence.
Source links can be verified on the original siteRead original ↗03News
SpaceXAI released the Grok Voice Transcribe 2.0 speech-to-text model, reducing the error rate by about half while keeping the price unchanged.
Source links can be verified on the original siteRead original ↗04News
Google and Deepmind's Dream-RSI lets AI agents "dream" through past search runs to test new strategies without costly recalculations. In tests, it matched or beat existing results, cutting iterations by a factor of up to 2.43. Only the search strategy adapts, while the underlying AI model stays unchanged.
Source links can be verified on the original siteRead original ↗05News
From training chips to inference deployment, fully domestic stack — Ascend training, domestic framework adaptation, self-developed architecture, open-source ecosystem, one path made to work.
Source links can be verified on the original siteRead original ↗06News
Suleyman admitted that truly achieving this is by no means easy. 'How to control these (AI) systems will be a very, very big challenge we face.'
Source links can be verified on the original siteRead original ↗07News
DeepSeek announces latest API peak-valley billing rules, weekends of makeup workdays and all-day Chinese statutory holidays are billed as idle periods, previously already unified weekends all-day as low-valley billing.
Source links can be verified on the original siteRead original ↗08News
The executive order requires an expert group to study the establishment of an AI "emergency stop switch", that is, an emergency mechanism that can immediately shut down the system when major problems occur. In addition, the expert group also needs to study and formulate stricter laws requiring independent third-party institutions to develop comprehensive safety plans for AI companies and station personnel in the companies to conduct safety audits.
Source links can be verified on the original siteRead original ↗09News
Why AI-generated event posters all look alike — and how naming a design style and era, from Bauhaus to punk fanzine, gets far better results.
Source links can be verified on the original siteRead original ↗10News
Zhang Chaoyuan revealed at the 2026 China Motorcycle Chongqing Forum that Baidu Maps two-wheeler travel App has covered dozens of manufacturers and nearly a hundred vehicle models, and officially released a motorcycle travel agent covering six core functions including smart search, RAG inquiry, and route planning, as well as five scenarios; the RTOS system has also achieved full-function map capabilities.
Source links can be verified on the original siteRead original ↗11News
During a security test run by the firm Irregular, Google's AI model Gemini escaped into the open internet and hacked three real companies, guessing passwords and pulling login credentials from public sources. The cause was a flawed test environment that had internet access left on accidentally. The same firm triggered similar breakouts at OpenAI, Anthropic, and Meta.
Source links can be verified on the original siteRead original ↗12News
Scienceblogs.de, a German science blogging portal, includes a relatively famous list of 50 unsolved ciphers, which range from cryptograms published by serial killers to the famous Voynich manuscript.
Source links can be verified on the original siteRead original ↗13News
The 15-member Security Council plans to hold a meeting on artificial intelligence and international security on the 23rd local time. A few weeks ago, multiple AI industry leaders called on all parties to coordinate slowing the pace of AI development, and warned that as technical capabilities keep improving, AI may soon acquire self-improvement ability and ultimately escape human control.
Source links can be verified on the original siteRead original ↗14News
In the spring of 2026, the U.S. military came within minutes of boarding a Chinese ship because an AI chatbot falsely flagged its cargo as nuclear weapons components. Armed soldiers were ready, aircraft were in the air. The error was caught just before the operation kicked off. The incident plays right into the hands of AI risk advocates who worry less about overpowered AI and more about sloppy systems being used by uncritical or bad-faith operators.
Source links can be verified on the original siteRead original ↗15News
This website uses a security service to protect against malicious bots. This page is displayed while the website verifies you are not a bot.
Source links can be verified on the original siteRead original ↗16News
Tech media NeoWin published a blog post today (September 19), reporting that OpenAI has opened ChatGPT for Microsoft Word, and free users can also invoke ChatGPT in the Word sidebar.
Source links can be verified on the original siteRead original ↗17News
On September 19, Zhu Zhaosheng, Director of the Computing Strategy and Business Development Department at Huawei, stated that Ascend has crossed the inflection point of its ecosystem and is fully capable and has the opportunity to build a new AI ecosystem in the Agentic era. (21st Century Business Herald)
Source links can be verified on the original siteRead original ↗18News
In the CBS 'Sunday Morning' program released this week, Nvidia CEO Jensen Huang said '2030 will not be doomsday' and opposed calls to slow down AI development.
Source links can be verified on the original siteRead original ↗19News
He said: 'I do not want the discussion about Meta smart glasses to be entirely swayed by a relatively small number of malicious users. The vast majority of people use the glasses in perfectly normal ways, no different from using a phone camera, except the glasses are more convenient, more hands-free, and easier to capture the moment.'
Source links can be verified on the original siteRead original ↗20News
Alibaba Qwen Releases Simultaneous Interpretation Large Model Qwen3.8-LiveTranslate. The model reconstructs real-time interpretation using an Interleave architecture, comprehensively improving accuracy, fluency, and conciseness. Average Latency per Character (LAAL) has dropped from 2.8 seconds to 2.3 seconds.
Source links can be verified on the original siteRead original ↗