CCTV1 — broadcast
20260820 113900 UTC
128 transcript segments
Google Cloud Speech-to-Text API (Chirp) + Gemini 2.5 Flash Non-Thinking
Data courtesy of The GDELT Project (https://www.gdeltproject.org/), from the Internet Archive TV News Archive.
Machine transcription. Treat it as a searchable index of what was broadcast, not a verbatim quotation record.

[00:02:02] Hello, friends and viewers, welcome to Focus Interview. At the 2026 World Artificial Intelligence Conference and the High-level Conference on Artificial Intelligence Global Governance,
[00:02:06] 会 及 人 工 智 能 全 球 治 理 高 级 别 会 议 开 目 市 上
[00:02:09] President Xi Jinping proposed to jointly build a fair and reasonable
[00:02:13] global artificial intelligence governance system. The first point is
[00:02:17] to adhere to open co-operation to drive innovation and development,
[00:02:20] and to encourage open source collaboration and sharing.
[00:02:23] Around this conference, domestic open-source large models were intensively launched.
[00:02:27] KMK3 with
[00:02:31] 28,000 units
[00:02:34] and Alibaba's Deepsick, etc.
[00:02:38] Open source is becoming the most prominent feature of China's artificial intelligence.
[00:02:44] July 27, 2026, the new era of artificial intelligence.
[00:02:45] 月27號人工能越新期
[00:02:50] The launch of KimyK3 is not the first open-source of a large Chinese model, but it can produce a chess piece of 28,000 digits.
[00:03:30] What's even more surprising is that this open-source not only includes
[00:03:34] the full weight of the model, but also technical reports and three underlying
[00:03:38] model training methods. What is open
[00:03:41] source? We can compare a large model to a restaurant. Using
[00:03:45] a closed-source model is like going to a restaurant to eat. The food is delicious, but you can't
[00:03:49] take it home. If you want to eat, you can only come to the restaurant. But using an open-source model is
[00:03:52] like sharing the secret of your restaurant. You can come to the restaurant to eat,
[00:03:56] or you can take the ingredients home and make it yourself. You can also.
[00:04:33] Kimi K3 was launched as an open-source project and immediately ranked first globally in the
[00:04:36] Fronten and Code
[00:04:37] Arena with 1679 points.
[00:04:44] At the World Artificial Intelligence Conference on July 17th, although there was only a simple video introduction,
[00:04:49] Kimi's booth still attracted many domestic and foreign visitors who came for technical exchanges or experiences.
[00:04:57] Kimi launched K3, such a big
[00:05:33] chip design animation made by K3.
[00:05:35] You might think K3 only made such an animation, but actually it's not. This is just K3 for chip design.
[00:06:32] Such a chip design task usually requires multiple experienced chip designers to collaborate, but K3 only received
[00:06:35] the simplest prompt, and independently completed the entire process.
[00:06:39] It can independently acquire these.
[00:06:54] The release of KMK3 is just one of the many new updates recently.
[00:07:00] Since July this year, China's open source has been upgraded intensively.
[00:07:06] On July 31st, Deepsiek released V4
[00:07:09] flash official version, its capabilities are greatly
[00:07:15] Deep seek V4 Pro official version.
[00:07:19] On August 3rd, 208 launched a large model
[00:07:23] 3.8
[00:07:23] max 24,000 units, then open source.
[00:07:29] it is the second largest open-weighted model in China
[00:07:32] so far, just behind K3 and according to
[00:07:35] Alibaba, it trails just behind anthropic on certain
[00:07:39] benchmarks on the arena leerboard and it even
[00:07:50] The spring report of 2026 shows that the download volume of open-source models developed in China
[00:07:54] accounts for 41% of the global total, surpassing the United States
[00:07:58] to rank first in the world. On the global mainstream large model invocation
[00:08:01] list, the top five ranked models all come from Chinese
[00:08:05] teams. In the past 12 months, domestic models have maintained
[00:08:09] the upper limit of the global open-source model scale for nine months.
[00:09:16] China is using
[00:09:18] its own way to achieve modernization.
[00:09:21] .
[00:09:26] Keep up the good work.
[00:09:27] .
[00:09:31] Openness and inclusiveness are the core attitudes of China's open-source large models.
[00:09:34] It is precisely this attitude that allows open-source large models to quickly enter
[00:09:38] enterprises, boost production, and transform into tangible and visible
[00:09:42] real productivity. Product testing is currently being conducted on this screen.
[00:09:46] In just a few seconds, defects on the
[00:09:50] product were identified by AI and a repair plan was provided,
[00:09:54] and the machine tool was directly controlled to complete the repair.
[00:09:57] In the past, this identification process was performed manually,
[00:10:01] which was less efficient. This AI system's technical foundation
[00:10:04] is called the Lan Jing Display Large Model,
[00:10:07] developed and designed by Jingdongfang, but its base model
[00:10:11] is from Cai Na.
[00:10:25] Mismatch, open-source models have this advantage.
[00:10:28] Our large model is deployed in our enterprise's domain. In this way, our
[00:10:32] information security and data security can be guaranteed to the greatest extent.
[00:10:36] Lower cost, deep customization, private deployment.
[00:10:39] These three advantages allow manufacturing enterprises like Jingdongfang
[00:10:43] to transform general open-source large models into industry-specific AI brains
[00:10:47] through secondary training for their own use. Currently, the Lan
[00:10:51] Jing Display Large Model has been implemented in Jingdongfang's nationwide
[00:13:25] How can such a good technology, or a breakthrough technology,
[00:13:28] be effectively used? This is a question that China has always been concerned about.
[00:13:32] Therefore, the views put forward by President Xi Jinping in Shanghai this time
[00:13:36] actually precisely address this concern. On July 17, 2026,
[00:13:40] President Xi Jinping
[00:13:44] delivered a keynote speech at the 2026 World Artificial Intelligence Conference and the High-level Conference on Artificial Intelligence Global Governance,
[00:13:47] 全 球 治 理 高 级 别 会 议 开 幕 市 上 发 表 主 指 讲 话
[00:13:51] proposing to jointly build a fair and reasonable global.
[00:13:57] The first point is to adhere to open co-operation,
[00:14:01] driving innovation and development. China advocates open cooperation
[00:14:05] and a governance concept for mutual development, which highly aligns with the development laws of the open-source ecosystem.
[00:14:09] Openness does not mean laissez-faire.
[00:14:13] The core is to establish a governance mechanism for the development of adaptive technologies,
[00:14:17] achieving a dynamic balance between innovative vitality and risk prevention and control.
[00:14:20] Beyond concepts, there are actions. As early as
[00:14:24] 2023, China took the lead in releasing the Global Artificial Intelligence Governance
[00:14:27] Initiative, clearly advocating a people-centered and AI for good governance philosophy.
[00:14:31] On July 17, 2026, China's Ministry of Industry and Information Technology,
[00:14:35] together with relevant parties, jointly guided the release of the International Artificial Intelligence
[00:14:39] Ethics Governance Action Plan, focusing on ethical governance throughout the entire lifecycle of artificial intelligence,
[00:14:42] risk classification and grading prevention and control, flexible
[00:14:46] governance mechanism construction and other directions, urging all countries to strengthen policy
[00:14:50] coordination and practical cooperation. On July 16,
[00:14:54] the establishment of the World Artificial Intelligence Cooperation Organization Agreement signing ceremony
[00:14:58] was held in Shanghai. In the next stage, the core question
[00:15:01] that truly needs to be answered is how to effectively identify,
[00:15:05] prevent, and mitigate new risks brought by artificial intelligence
[00:15:08] while promoting technological innovation, and to explore a dynamic balance between development and security.
[00:15:12] Roughly speaking, the answer can be approached from three directions:
[00:15:16] first, promoting the coordinated development of innovation and security governance.
[00:15:19] Second, how to implement classified and graded
[00:15:23] approaches based on risk levels.
[00:15:38] From domestic open-source large models deeply empowering various industries, to China
[00:15:42] proposing to build a fair and reasonable global artificial intelligence governance system
[00:15:45] with four points of view, China is demonstrating through practical actions
[00:15:48] that the future of artificial intelligence cannot be separated from the power of global
[00:15:52] open collaboration. For the world,
[00:15:55] only by upholding open co-operation and promoting development and security,
[00:15:59] can artificial intelligence better serve all
[00:16:02] humanity. Thank you for watching today's Focus Interview. Goodbye.
[00:16:05] .
[00:16:38] News background analysis.
[00:16:39] .
[00:16:42] In-depth analysis of social hot spots.
[00:16:44] .
[00:16:47] Commentary on popular topics.
[00:16:49] .
[00:16:52] Please watch Focus Interview.
[00:16:54] .
[00:21:50] Village development, everywhere vibrant.
[00:21:53] .
[00:21:56] Beautiful villages like paintings, struggle. .