Baichuan Intelligence has released the Baichuan2-192K large model. Its context window length reaches up to 192K, claiming to be the longest context window in the world. iFlytek has released the Spark Medical large model, which comprehensively surpasses GPT-4 in medical massive knowledge Q&A, complex language understanding, professional medical text generation, and diagnostic and treatment recommendation answer rate.

Image source: Generated by Wujie AI

Funding

Google commits to provide $2 billion in funding to AI startup Anthropic

According to The Wall Street Journal on October 28, Google has committed to provide $2 billion in funding to the AI startup Anthropic. Sources said that Google has pre-invested $500 million in this OpenAI competitor and agreed to add another $1.5 billion over time.

AI startup Giga ML completes $3.6 million seed round funding, with participation from Y Combinator

According to Financialexpress on October 23, AI startup Giga ML has secured $3.6 million in seed round funding, led by venture capital firm Nexus Ventures Partners, with participation from Y Combinator, Liquid 2 Venture, 8vdx, and Y Combinator's President and CEO Garry Tan as angel investors.

Databricks acquires data company Arcion for $100 million, plans to integrate its technology into MosaicML

According to CNBC on October 23, analytics company Databricks CEO Ali Ghods stated that the company has agreed to acquire enterprise data company Arcion for $100 million, including incentives. Ghodsi said the company plans to integrate Arcion's technology into MosaicML, which will serve as "a data source for Mosaic when customers want to build their own AI models using the tool."

German AI startup Oxolo secures €13 million in funding to develop generative video platform

According to TechCrunch on October 24, German AI startup Oxolo has completed a €13 million Series A funding round led by DN Capital. German angel investors participating in this round include Christian Vollmann, Johannes Plehn, Thomas Hagemann, Stefan Wiskemann, Alex Täubert, and Mischa Rürup. The funds will be used to develop an AI-driven generative video platform.

Brilliant Labs raises $3 million in funding to develop AI-based AR glasses

According to Zhanzhangzhijia on October 25, Brilliant Labs recently secured $3 million in funding to develop its AI-based augmented reality glasses. This funding will help the company promote the development of generative AI technology, strengthen its position as an AR device provider, and build an open-source ecosystem for developers and consumers seeking reliable personal AI assistants.

AI voice service provider Wiz AI raises $30 million in Series A+ funding

According to IT Orange on October 25, AI voice service provider Wiz AI, aimed at helping call centers undergo intelligent transformation and providing a full range of solutions, recently secured $30 million in Series A+ funding from Wavemaker Partners, Cloud Angel Capital, Hillhouse Capital, New Horizon Capital SingTel Innov8, Insignia Ventures Partners, K3 Ventures, and Cloud Angel Capital.

UK AI cybersecurity startup Harmonic raises $7 million in funding

According to Zhanzhangzhijia on October 24, UK AI startup Harmonic Security recently secured $7 million in funding to build technology to safeguard enterprise generative AI deployments. Harmonic will develop software to address the issue of unregulated AI applications collecting large-scale corporate data.

AI startup CentML raises $27 million in funding, with participation from Nvidia

According to TechCrunch, on October 25, CentML announced that it has raised $27 million in an expanded seed round of funding, with participation from Gradient Ventures, TR Ventures, Nvidia, and Microsoft Azure AI Vice President Misha Bilenko. CentML Co-founder and CEO Gennady Pekhimenko stated that the new funds will be used to strengthen CentML's product development and research efforts, as well as expand the startup's engineering team and its 30-person workforce distributed in the United States and Canada.

AI software developer Moreh completes $22 million in Series B funding

According to TechCrunch on October 26, AI software developer Moreh has completed a $22 million Series B funding round, led by AMD and South Korean telecom company KT. Moreh stated that its flagship AI software MoAI is similar to Nvidia's CUDA, but compatible with existing machine learning frameworks (such as Meta's PyTorch, Google's TensorFlow) and applications and AI models that previously could only run on Nvidia.

AI startup Credal.ai completes $4.8 million in seed round funding

According to TechCrunch on October 26, Y Combinator-backed startup Credal.ai has completed $4.8 million in seed round funding, led by Spark Capital. The company aims to provide enterprises with a method to connect their internal data to text generation and cloud-hosted AI models.

AI enterprise Cranium incubated by KPMG raises $25 million in funding

According to VentureBeat on October 26, AI enterprise Cranium, incubated by KPMG, announced a $25 million Series A funding round. This round was led by Telstra Ventures, with participation from KPMG LLP and SYN Ventures, bringing Cranium's total funding to $32 million.

Large Models

iFlytek: Officially releases the Spark Cognitive Large Model V3.0, surpassing ChatGPT overall

According to Babite News on October 24, the 2023 iFlytek Global 1024 Developers' Festival opened in Hefei. iFlytek announced the official release of the iFlytek Spark Cognitive Large Model V3.0. It is understood that the Spark large model has overall surpassed ChatGPT.

Baichuan Intelligence releases the Baichuan2-192K large model, with a context window length exceeding GPT-4 by about 14 times

According to DoNews on October 30, Baichuan Intelligence has released the Baichuan2-192K large model. Its context window length reaches up to 192K, claiming to be the longest context window in the world. Baichuan2-192K can process about 350,000 Chinese characters, which is 4.4 times that of the best large model Claude2 (supporting 100K context window, tested at about 80,000 characters), and 14 times that of GPT-4 (supporting 32K context window, tested at about 25,000 characters).

Zhipu AI releases self-developed third-generation base large model ChatGLM3

According to "Science and Technology Innovation Board Daily" on October 27, Zhipu AI today released its self-developed third-generation base large model ChatGLM3. Zhipu AICEO Zhang Peng introduced that ChatGLM3 has achieved iterative upgrades with several new features, including the multimodal understanding capability CogVLM—achieving SOTA effects on more than 10 international standard image-text evaluation datasets, significantly enhancing semantic and logical capabilities, and speeding up 2-3 times compared to the current best open-source implementation vLLM.

Tencent's Hybird large model's Chinese capability overall surpasses GPT3.5, with a 20% improvement in coding capability

According to the Economic Daily News on October 26, Tencent's Hybird large model has undergone a new upgrade, with its overall Chinese capability surpassing GPT3.5, and its coding capability significantly increasing by 20%. Regarding the research and development cost of large models, Kang Zhanhui, the algorithm head of Tencent's machine learning platform, stated that Tencent currently has two directions for developing large models: for general large models, Tencent prioritizes effectiveness over cost investment and emphasizes long-term investment; for small models, the goal is to be more refined, applying different models to different scenarios while considering the cost-effectiveness and benefits. In simple terms, it's "go big with big models, go small with small models." Babite News reported on October 30 that Kunlun Wanwei officially announced the open source of its billion-level large language model "Skywork-13B" series, along with a matching open source Chinese dataset of 600GB and 150B tokens. The Skywork-13B series from Kunlun Wanwei currently includes two large models with 13 billion parameters: the Skywork-13B-Base model and the Skywork-13B-Math model. Additionally, the Skywork-13B series of large models from Kunlun Wanwei is set to be fully open for commercial use—developers can use it for commercial purposes without the need for application. According to a report from the Shanghai Securities News, on October 22, at the "Global First Commodity Trade Domain Voice Model" launch event, Xiaoshangpin officially released the world's first large language model in the commodity trade domain and the newly upgraded Chinagoods AI innovation service platform. It aims to use AI digital humans, AI mapping, AI translation, and intelligent customer service in the trade domain to find new growth for small commodity trade. Stability AI Japan recently released two Japanese language models, named "Japanese Stable LM3B-4 E1T" and "Japanese Stable LM Gamma 7B," as reported by Zhanzhangzhijia on October 25. The former has approximately 30 billion parameters, while the latter has a massive scale of up to 70 billion parameters. These two models are released under the Apache 2.0 license, allowing commercial use and bringing a revolutionary breakthrough to Japan's natural language processing field. On October 26, Jina AI announced the launch of the jina-embeddings-v2 model, claiming to be the only open-source product that supports a context length of 8K (8192 tokens) and is functionally and performance-wise similar to OpenAI's text-embedding-ada-002, as reported by IT Home. Regarding applications, iFlytek released the Spark Medical large model, which comprehensively surpasses GPT-4 in medical aspects, as reported by the Securities Times on October 24. Additionally, the Honor Magic 6 will be equipped with the Qualcomm Snapdragon 8 Gen 3 mobile platform, supporting a 70 billion parameter AI edge-side large model, as reported by Pacific Computer Network on October 25. Furthermore, Xiaomi's Penglai OS will integrate AI large models into the system, becoming the first to support NPU deployment, as reported by IT Home on October 26. Additionally, Kuaishou is testing "AI Xiaokuai," which can use the Kuaiyi large model to provide intelligent replies in the comment section, as reported by 36Kr. Shutterstock announced the launch of an AI image editor for its 7.5 billion image library, as reported by VentureBeat on October 26. In the realm of major companies, Microsoft announced its plan to invest nearly $32 billion in Australia over the next two years to expand its cloud computing and AI infrastructure, as reported by Jiemian News on October 23. Additionally, Google Maps announced a series of AI feature updates, including support for "immersive views," as reported by IT Home on October 26. Furthermore, Intel's CEO stated that sales channels for Gaudi and AI accelerator chips have nearly doubled in the past 90 days, as reported by Jijin News on October 27. In terms of viewpoints, Baichuan Intelligence CEO Wang Xiaochuan expressed that large models transform language into mathematical models, allowing the understanding of underlying rules, as reported by Jiemian News. Additionally, an early investor in OpenAI stated that most AI companies find it difficult to be profitable.According to the report from Zhanzhangzhijia on October 30, early OpenAI investor Vinod Khosla stated at a recent tech conference to the Financial Times that the investment frenzy in the AI field following the launch of OpenAI's ChatGPT may not lead to similar success. He pointed out that currently, most AI venture capital investments may result in losses. He compared this year's AI hype to last year's cryptocurrency investment frenzy, and mentioned that many latecomers are just following the trend because others are investing, which is exactly what is happening in the AI field. Three Turing Award winners jointly published a paper titled "Managing AI Risks in an Era of Rapid Development" recently, as reported by Pingwest on October 30. The three scientists, Hinton, Bengio, and Yao, expressed in the paper that when advanced AI technology is used correctly and fairly, it can help solve long-standing problems such as diseases, poverty, and the environment. However, at the same time, powerful AI technology may also bring huge risks. The three scientists once again called for global efforts to regulate the development and application of autonomous AI through technology, policies, and legal means to avoid a future dominated and misused by AI. In a report from Bloomberg on October 23, Morgan Stanley analysts, led by Matthew Cost, wrote that AI tools could reduce the costs of AAA games by as much as 15%. Gaming platforms including Microsoft's Xbox and Sony's PlayStation are expected to be the "most obvious beneficiaries of the gaming industry adopting AI." Additionally, these companies include Unity, Roblox, and Tencent.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。