AI Card Bao 173.208.162.207
Home / Blog

BLOG / INTELLIGENCE

Blog

370 articles, updated continuously.

2026.11.16

Is it more effective to use a series of prompts to make Claude first act as five advisors and have them argue with each other before answering the question, rather than asking directly?

These past couple of days, a certain prompt has been promoted everywhere under the banner of “Claude’s brand-new feature.” But let me cool your enthusiasm right away: this isn’t actually a new feature at all; it’s just a set of instructions that were already included in the project documentation and can be used in any chat conversation. Nevertheless, the approach behind it is worth discussing. It allows the model to act as five different “advisors” before answering each question…

Industry News 52 reads
2026.11.16

The Gemini 4 Carbon model has been integrated into the international version of Tencent WorkBuddy. Pro users can now take advantage of its extensive contextual features, which include information pertaining to millions of different contexts.

On November 15th, Tencent's AI-powered office product WorkBuddy International version was integrated with Google's Gemini 4 Carbon. Currently, this feature is only available to users subscribed to the Pro plan. Gemini 4 Carbon represents Google's latest advancement in long-term intelligent...

Industry News 53 reads
2026.11.16

Helen Turner, who voted to remove Ultraman back then: I would do the same again if given the chance.

Among the four votes cast by the OpenAI Board of Directors in 2023 to remove Otman from his position, one of them came from Helen Turner. A few days later, Otman returned to his post, and Helen Turner left the board. Three years later, during an episode of The Verge’s Decoder podcast, the host asked her what she would do if given the chance to do it all over again…

Industry News 53 reads
2026.11.16

Tencent’s Hy4flash has been launched: It can process 12 billion operations per second and decode data at a rate of over 150 terabytes per second. Moreover, the cost per single task is only one-twentieth of that of Hy4.

Tencent’s Hunyuan released Hy4flash today. The official description for this model is quite straightforward: it is currently the most cost-effective Hy model, specifically designed for everyday tasks. Its technical specifications include a total of 224B in parameters (of which 64B are dedicated to Engram parameters), while the number of activation parameters amounts to only 12B. It supports…

Industry News 52 reads
2026.11.16

OpenAI has launched a scientific blog, and the first article in this series explains how physicists use GPT-6 Astra to advance lattice theory.

OpenAI has launched a new scientific blog, and the first article was written by physicist Nathaniel Craig. In it, he describes how he used GPT-6 Astra to propose a candidate lattice definition for the standard model. This development has generated quite a stir within the physics community…

Industry News 50 reads
2026.11.16

Apple’s first smart glasses, codenamed N50: no screen at all, relying solely on Siri; scheduled to be unveiled at WWDC 2027.

Apple’s smart glasses now have a more specific timeline: they are scheduled to be unveiled at next year’s WWDC in June, with the goal of starting shipments by the end of the year. If there are delays in software development, this schedule may change. The internal code for these devices is N50. The first generation of these smart glasses did not come equipped with a display screen; all interactions were handled entirely through Siri, which was capable of recognizing the environment surrounding the user…

Industry News 47 reads
2026.11.16

Amodei in conversation with Damon: How can the commitment of 900 billion dollars in computing power be fulfilled? And why does Anthropic refuse to hire people who are “technically capable but difficult to work with”?

One week before its listing, Anthropic’s CEO Dario Amodei went to New York to have a public discussion with Jamie Dimon of JPMorgan Chase. Dimon asked very direct questions. Anthropic had disclosed a specific figure: the total value of its future commitments related to computing power, leasing services, and data centers amounts to approximately …

Industry News 53 reads
2026.11.16

The United States, Japan, the United Kingdom, Australia, and other countries signed the San Francisco Declaration: They compiled a list of key AI technologies and supply chains in order to prevent their adversaries from obtaining them.

The United States, together with a group of its allies, signed a new AI agreement whose objective is clear: to counter China. The initial signatories include Japan, Australia, Israel, and the United Kingdom. The document is titled “The San Francisco Declaration on the Security Foundations for Sustainable Prosperity.” U.S. Undersecretary of State for Economic Affairs Herbert Steinberg said that South Korea…

Industry News 52 reads
2026.11.16

The Kimi K3.1 Pro has been migrated to the Qwen4-Max platform: its listed price is now half lower, yet according to third-party tests, the cost of each individual question has actually increased by 55%.

I currently have a set of agents that run on the Kimi K3.1 Pro platform. When I saw the price list for Qwen4-Max today, my first thought was: “With a cost of $1.5 to deploy and $4.5 to operate, it’s definitely much cheaper than Kimi’s options, which require $2 or $10 respectively…”

Industry News 51 reads
2026.11.16

Machine dogs equipped with rifles were deployed in actual combat operations in Haiti; they resembled the YuShu Go2 in appearance.

El Salvador’s President Bukele released a video showing how robotic dogs, equipped with rifles, patrolled a stadium alongside soldiers during a joint operation between El Salvador and Haiti. That stadium had been under the control of gangs for over two years before it was reclaimed this time. The robotic dogs in the video closely resemble Yuzo’s Go2 models; they have…

Industry News 49 reads
2026.11.16

LongCat-2.5 was trained entirely using the Ascend platform. On the same day, a study indicated that six domestically developed models still retained the “Blackwell fingerprint” characteristic associated with their training methodology.

The Meituan LongCat team released the weights for LongCat-2.5 one week after its publication. The license is licensed under the MIT license and can be used for commercial purposes. This model is designed for agents that are capable of executing long-duration tasks independently, with programming being a key aspect of its functionality. Each token activates approximately 48 billion parameters…

Industry News 50 reads
2026.11.16

The beta version of Telegram for iOS contains a 3D duck – this feature might represent the precursor to the AI assistant that will be available in future chats.

In the debugging menu of the Telegram iOS 13.3 beta version (build 37388), there is a feature called “Companion”: a 3D duck will appear in the private chat window. For now, this feature doesn’t do anything—it doesn’t reply to messages or engage in any interaction; it simply stays there…

Industry News 50 reads