Anthropic this week filed a preliminary S-1 prospectus of more than 300 pages, the first time the market has been able to fully see the fundamentals of a pure AI lab. After the prospectus was disclosed, revenue, compute costs, corporate governance and pricing remain the areas analysts are watching most closely.
#Industry news#Anthropic
Sources familiar with the matter say Nvidia is in talks to provide AI cloud service provider Nscale with a multi-billion-dollar debt credit guarantee for the construction of a 460-megawatt data center in West Virginia; the facility will be equipped entirely with Nvidia Vera Rubin GPUs and leased to Anthropic. The agreement takes the form of a residual-value guarantee on the data center assets themselves, rather than on the lease or the chips, which could allow Nscale to obtain investment-grade financing; if Nscale fails to pay construction costs on schedule, Nvidia will pay on its behalf and temporarily take over the facility. Nscale previously disclosed in its prospectus that it had signed a $45 billion contract with Anthropic, and Anthropic also plans to sign for the second phase of the campus.
#Industry news#English text is being prepared.#English text is being prepared.
TechCrunch has put together a last-minute scramble list for tech enthusiasts who haven't prepared a Halloween costume, including looks such as "recursive self-improvement," a cute-looking AI agent that asks for read-write access to your Gmail, a passerby secretly recording others while wearing Meta glasses, a Claude token usage limit notification, and the PHP vulnerability that AI dug up last week. Most of these costumes only require existing clothes, cardboard, or stickers, and the article also reminds readers not to actually film others without consent.
#English text is being prepared.
According to a Financial Times report, Blackstone told investors participating in the $63 billion Anthropic chip financing deal that it expects Anthropic to receive a "high" junk credit rating after its IPO, specifically in the BB range. Blackstone had previously said Anthropic would obtain an investment-grade rating by 2027. Anthropic formally submitted its listing application this week; TechCrunch contacted Anthropic and Blackstone about this, and both declined to comment.
#Industry news#Anthropic
Intel shares plunged about 28% on Friday, one of its worst single-day performances, after fourth-quarter guidance came in weaker than expected, shrinking the value of the government's stake by $12 billion. The company expects fourth-quarter revenue of $15 billion to $16 billion, below analysts' expectations of about $17 billion; gross margin will fall from 43% to 37%, and it expects negative free cash flow. CEO Lip-Bu Tan said customers have signaled a PC slowdown due to memory prices and supply issues, with system end prices rising noticeably during the year; data center revenue grew 78% year over year, but capacity expansion is constrained, and the next factory in Arizona is not expected to come online until 2028.
#Industry news#English text is being prepared.
The ElevenLabs blog compared the differences between warm transfer and cold transfer in call centers: warm transfer provides the receiving party with background information about the caller before connecting, while cold transfer connects directly without prior notice. The article recommends warm transfer for complex or sensitive issues and cold transfer for simple, high-frequency routing, and notes that ElevenAgents supports three transfer types, with Conference Transfer on its native Twilio integration able to pass warm transfer information to human agents.
#Tutorials / practice#Speech#Agents
Inspired by Divyansh Agrawal's qwenputer, Drew Breunig wrote a runnable NES 6502 CPU emulator into the weights of Qwen3.5-4B, and it can now (very slowly) run NES ROMs such as Super Mario Bros inside Qwen. In a Hacker News comment, he disclosed the prompts used when building the system with Opus 5.5, noting that Opus found and used the NES CPU consistency test suite that Christopher Pow published 15 years ago.
#English text is being prepared.#English text is being prepared.#Open-source ecosystem
At the Young Scientists Conference of the 2026 World Laureates Forum, top scientists including Michael Levitt, Laurent Lafforgue, and Makoto Fujita responded to questions such as whether AI is squeezing young people's room for development and the reversal of the teacher-student relationship. Levitt said he has been using AI for nearly four years and believes that using AI is essentially enhancing the user's own intelligence; Fujita judged that for at least the next decade, AI will remain merely a tool. Jürgen Caro believes that the only remaining value for professors is critical thinking, while Ma Jianpeng reminded students that using AI to expand papers is padding them with fluff.
#Industry news#English text is being prepared.
At the 2026 World Laureates Forum, scientists including Novoselov and Chen Yunji discussed AI empowering scientific research: AI can modify the atomic structure of two-dimensional materials and predict new functions, and robotic laboratories form a closed loop with large models; Chen Yunji's team produced the AI-designed "Qimeng 1" in 2021, and in 2025 "Qimeng 2" reached the performance level of mainstream mobile phone CPUs from more than a decade ago. Fields Medal winner Laurent Lafforgue warned that the danger of learning with AI is that knowledge remains superficial.
#Industry news#English text is being prepared.#Embodied AI
At the AI parallel forum of the 2026 World Laureates Forum, experts including Turing Award winner Martin Hellman, Wolf Prize in Mathematics winner Noga Alon, and Chinese Academy of Sciences academician Chen Yunji pointed out that AI safety cannot be solved by technology alone. Chen Yunji believes that mainstream large language models are based on statistical principles, and their accuracy can never reach 100%; this 1% uncertainty is enough to cause harm, and he advocates that every large model should come with a built-in ethical review mechanism. He Jifeng, meanwhile, emphasized that AI must be explainable, verifiable, and traceable, and aligned with human values.
#Industry news#Safety / alignment#English text is being prepared.
Embodied intelligence means empowering AI onto physical entities, enabling models to rely on environmental perception for self-iterative learning, which is fundamentally different from traditional robots driven by preset programs. The humanoid form is not a bionic gimmick, but an engineering choice to adapt to existing human physical environments such as door handles, stairs, and tools. Wang Xingxing, founder of Unitree Robotics, believes that the "ChatGPT moment" of embodied intelligence—completing 80% of tasks through natural instructions in 80% of unfamiliar scenarios—could come in as soon as 2–3 years, or as long as 5–10 years.
#English text is being prepared.#Embodied AI
Muse posted "dress for the job you want," along with an image, without further explanation.
#English text is being prepared.
Even the world's best drivers like being driven
#Industry news#English text is being prepared.
BREAKING: Cathie Wood says SpaceX is severely undervalued. SpaceX brings together AI, robotics and energy storage. But traditional space, cloud and AI analysts work separately, making it harder to understand how these technologies strengthen one another. Wood says firms are now assigning teams of analysts to SpaceX. As they collaborate and compare notes, she expects that undervaluation to clear up. https://t.co/octmFBgmcm
#English text is being prepared.
#Papers/Research#Reasoning#English text is being prepared.
#Industry news#English text is being prepared.#English text is being prepared.
#English text is being prepared.
#Tutorials / practice#Agents#English text is being prepared.
as predicted https://t.co/rJ2gFuxgcj
#English text is being prepared.#OpenAI
Baseten Base Labs' toy experiments show that at the same number of steps, the PPO critic performs on par with or behind GRPO, while a "perfect critic" can eliminate at most 40-60% of the advantage variance. The authors argue that a value function is only worthwhile when the gradient noise it eliminates exceeds its training cost; the only scenario where it might help is with a small value head, but even then the gains are only marginal.
#Papers/Research#English text is being prepared.#Reasoning