- AI Breakfast
- Posts
- The State of AI in China
The State of AI in China
And an AI tool to create business workflows
Good morning. It’s Monday, May 15th.
We’ll take a look at the global AI race between the US and China, and ChatGPT releases 70 new plugins for Plus users.
In today’s email:
State of China’s AI
ChatGPT Plugins
Scribe AI: Creating Business Workflows
Trending AI Tools
This Week’s AI Research Papers
You read. We listen. Share your feedback by replying to this email, or DM us on Twitter.
Also, did you know we have a podcast? Check it out here on Apple Podcasts and Spotify.
The State of AI in China: Government Regulation, US Bans, and a Trillion Dollar Pledge
The quest for AI supremacy has intensified on the global stage, with the US and and China poised to dominate the race - though sanctions and regulations may delay the blossoming of China’s abilities.
China's rapid advancements in AI research and development have raised concerns that extend far beyond the pursuit of dominance. The potential weaponization of AI against adversaries and its impact on the rights of Chinese citizens add weight to these concerns, further amplified by China's remarkable annual output of STEM graduates, dwarfing its American counterparts by a factor of ten.
Regulatory Measures: Navigating the Maze
China has recently proposed draft measures aimed at regulating generative AI services developed within its borders.
The proposed regulations demand that AI-generated content be "true and accurate," presenting a challenge for large language models programmed to generate "likely" strings of characters rather than "truthful" ones.
Navigating the complexities of the regulatory environment while ensuring adherence to political censorship will shape the direction of AI development in China, which will likely specialize in a more task-specific proprietary AI model approach, as opposed to a user-friendly language model environment that will flourish in the US.
Implications of US Ban and China's Trillion-Dollar Pledge
The US ban on the A100, a widely used NVIDIA GPU in AI applications, and China's “trillion-dollar pledge” to AI are pivotal factors shaping the state of AI within China.
The reliance on advanced chips from US designer Nvidia has been severely hindered by the ban on A100 chip sales to China, forcing companies to resort to slower alternatives like the A800, which are only capable of 1/3 of the computing power of the A100. Nevertheless, Chinese companies continue to make strides in AI, with estimated progress trailing no more than two years behind the pace set by OpenAI.
Although China lacks a clear counterpart to OpenAI at present, major Chinese companies such as Alibaba, SenseTime, and Baidu have introduced generative AI products, with Baidu's AI bot, Ernie, leading the pack - though it is currently far behind the capabilities of ChatGPT, and far more censored.
Competition and Challenges in the Chinese Market
Tencent and ByteDance (TikTok) will soon emerge as formidable competitors in China's AI market, leveraging their vast data reserves and extensive customer bases. These companies must navigate Beijing's strict AI regulations, which hold them accountable for any "incorrect" answers generated by their potential AI systems. Will these restrictions apply to the international deployment of AI systems? That remains to be seen.
Concerns surrounding China’s AI advancements and the potential consequences will intensify over the coming months, especially as tensions with Taiwan continue and the need for the world’s fastest GPUs, which are all made in Taiwan, becomes more paramount.
For a deeper dive into the players in China’s AI race, check out AI Superpowers: China, Silicon Valley, and the New World Order by Dr. Kai-Fu Lee, a former Apple, Microsoft, and Google executive (yes, all three).
This week’s sponsor:
Scribe AI
Scribe Al takes care of the most dreaded part of your job: documenting processes.
Scribe Al automatically creates SOPs, training manuals, new user guides and process overviews for any business process.
Scribe is your GPT-4-powered process documentation platform that automatically creates SOPs, help centers, new user guides and process overviews for any business process.
Scribe AI auto-generates step-by-step guides complete with screenshots and text by capturing your screen while you click and type.
Using your guides, Scribe AI can create full process documentation (including headings, subheadings, and detailed text) with your guides automatically embedded. No more staring at a blank document thinking, "ok, I have to teach someone how to do this, where do I start? what are all the steps". Scribe AI does it for you.
Thank you for supporting our sponsors
ChatGPT Plugins Are Available Now
OpenAI has announced the release of 70 new plugins for ChatGPT Plus users, covering a range of applications from entertainment to shopping to weather forecasts.
The plugins and web browsing capabilities are designed to respond to specific user prompts that require access to the web or specific applications, and is often referred to as “The App Store” for ChatGPT.
Developers interested in creating ChatGPT plugins can join a waitlist. Once approved, OpenAI documentation will guide the development process.
Trending AI Tools
Taskade.com: An automatic workflow generation tool using AI to create hundreds of workflow improvement opportunities for teams and solo operators
PrivateGPT: Interact with your data in an execution only environment without using the internet, substantially reducing data security/vulnerability issues created via connection to the web using a combination of LangChainAI + llama.cpp
Stockimg.ai: Easily design logos in minutes
Mage.space: Simple text-to-image AI generator
Tutorai.me: A tutoring AI platform for quickly setting the groundwork for learning about a wide variety of topics
Validatorai.com: Generate and primitively validate business ideas using AI
Top AI Research Papers This Week
Note: arXiv is a free online library where scientists share their research papers before they are published.
The paper proposes a new approach to autoregressive transformer modeling that addresses the challenges of training and using large models for very long sequences. The model uses a multiscale architecture that decomposes long sequences into smaller patches allowing it to use smaller models and less computation.
This new text-to-speech system combines advances in image generation with traditional TTS techniques. The system called TorToise, uses a denoising diffusion probabilistic model (DDPM) to generate high-quality video waveforms from text descriptions.
The author argues that we must think of strategies “beyond traditional AI alignment.” Instead of stifling an AI’s cognitive capacity to align it with human values, we should encourage a sense of agency and ownership in the AI. Establishing a sense of personhood and embodiment can achieve better alignment than mere control and suppression.
This paper surveys 51 leading experts from AGI labs, academia, and civil society to identify best practices in AGI safety and governance. The survey found that there is a broad consensus among experts that AGI labs should implement a wide range of safety and governance practices including conducting pre-deployment risk assessments, evaluating models for dangerous capabilities, commissioning third-party model audits, and imposing safety restrictions on model usage.
The authors of the paper identify three central challenges to applying AI to wireless receivers: limited computational and memory resources, dynamic channels, and the need for AI receivers to operate in a variety of scenarios.
They review approaches to address these challenges, including designing compact AI architectures, using self-supervised learning to generate training data, and developing training algorithms that are robust to dynamic channels.
PALR combines user behavior data with LLMs to generate user-preferred items. First, it uses user/item interactions as guidance for candidate retrieval. Then it adopts a LLM-based ranking model to generate recommended items. PALR fine-tunes a 7 billion parameters LLM for the ranking purpose. It takes retrieval candidates in natural language format as input, with instruction which explicitly asks to select results from input candidates during inference.
3x the information, for less than $2/week
Stay informed, stay ahead: Your premium AI resource.
AI Breakfast Business Premium: a comprehensive analysis of the latest AI news and developments for business leaders and investors.
Email schedule:
Monday: All subscribers
Wednesday: Business Premium
Friday: Business Premium
Business Premium members also receive:
-Discounts on industry conferences like Ai4
-Discounts on AI tools for business (Like Jasper)
-Quarterly AI State of the Industry report
-Free digital download of our upcoming book Decoding AI: A Non-technical Explanation of Artificial Intelligence
Thank you for reading today’s edition.
Your feedback is valuable.
Respond to this email and tell us how you think we could add more value to this newsletter.
Read by employees from