BAIDU’s ERNIE 4.0: Diffusion Design using Ideogram by Bogdan Iancu

Baidu Introduces ERNIE 4.0 Foundation Model, Pioneering the Next Generation of AI-Centric Applications

ERNIE 4.0

Features:

  1. Upgraded Capabilities: ERNIE 4.0 boasts significantly enhanced performance in understanding, generation, reasoning, and memory.
  2. ERNIE Bot: Demonstrates the four core AI capabilities of ERNIE 4.0, including understanding complex human requests, generating content (text, images, videos), reasoning through problems, and memorizing and integrating incremental inputs.
  3. Performance Improvement: Since its beta testing in September, ERNIE 4.0’s overall performance has improved by almost 30%.
  4. Integration with Baidu Apps: Baidu has integrated ERNIE 4.0 into its suite of applications, solutions, and products, such as Baidu Search, Baidu GBI, Infoflow, Baidu Wenku, Baidu Drive, and Baidu Maps.
  5. Plugin Matrix AI Platform: Enables developers and enterprises to build AI plugins with low-threshold access and productivity tools.
  6. Qianfan Foundation Model Platform: A one-stop enterprise-level foundation model platform that consolidates a range of foundation models, datasets, and toolchains.

Benefits:

  1. Enhanced AI-native Applications: The introduction of ERNIE 4.0 has led to the creation of AI-native applications that offer improved productivity and creativity across various usage scenarios.
  2. Improved Search Experience: With the integration of Generative AI, Baidu Search can now provide more in-depth information from a single search, offering a next-gen search experience.
  3. Business Analytics: Baidu GBI, powered by Generative AI, accelerates the process of business analytics and decision-making.
  4. Efficient Content Production: Baidu Wenku, with foundation model capabilities, assists users in content production, reducing the time spent in tasks like preparing a keynote speech.
  5. AI-native Map Product: Baidu Maps, with integrated AI capabilities, offers a seamless travel experience by suggesting the next steps of action to users during their trip.

Other Technical Aspects:

  1. Generative AI Integration: Baidu has adopted a generative AI approach to rebuild its family of apps, solutions, and products.
  2. AI Cloud Strategy: Baidu AI Cloud aims to provide infrastructure such as the Baidu Qianfan Foundation Model Platform to empower enterprise customers in developing AI applications.
  3. AI in Intelligent Devices: Baidu introduced a series of smart products built on AI in intelligent devices, including Tiantian Home Robot, Xiaodu Qinghe Learning All-in-one Machine, and Smart Speaker Series Tiantian Casa.
  4. Autonomous Driving: Baidu’s Apollo Go, the world’s largest autonomous driving ride service platform, has provided over 4 million cumulative rides to the public.

In conclusion, ERNIE 4.0, launched by Baidu, represents a significant leap in the realm of AI-native applications, offering enhanced capabilities and integrations that promise to revolutionize various sectors and user experiences.

Other AI News

  • Nvidia Expands Generative AI Capabilities with TensorRT-LLM SDK

Nvidia has unveiled its plans to bolster support for its TensorRT-LLM SDK, extending it to Windows and integrating models such as Stable Diffusion. The primary objective is to enhance the performance of large language models (LLMs) and associated tools. TensorRT is designed to accelerate inference, which involves processing pre-trained data to generate results, such as creating a new Stable Diffusion image.

By introducing TensorRT-LLM, Nvidia aims to optimize LLMs like Meta’s Llama 2 and other AI models, such as Stability AI’s Stable Diffusion, enabling them to operate more swiftly on Nvidia’s H100 GPUs. The company highlighted that this acceleration would notably enhance the user experience for advanced LLM applications, including writing and coding assistants.

Nvidia’s strategy is twofold: to continue supplying the GPUs that train and operate LLMs and to offer software that boosts model performance, eliminating the need for users to seek alternative cost-effective generative AI solutions. The TensorRT-LLM will be publicly accessible, with the SDK available on Nvidia’s website.

Currently, Nvidia dominates the market with its powerful chips designed for training LLMs, such as GPT-4. The soaring demand for its H100 GPUs has pushed estimated prices to a staggering $40,000 per unit. Furthermore, Nvidia has teased the launch of a new GPU, the GH200, slated for release next year. This robust performance is reflected in Nvidia’s impressive revenue surge, reaching $13.5 billion in the second quarter.

However, the dynamic landscape of generative AI is witnessing rapid changes. Alternatives to run LLMs without the need for high-cost GPUs are emerging. Tech giants like Microsoft and AMD are venturing into creating their chips, reducing dependency on Nvidia. AMD has expressed intentions to acquire software firm Nod.ai to optimize LLMs for AMD chips. Meanwhile, companies like SambaNova are introducing services that simplify model operations.

While Nvidia continues to lead in the generative AI hardware domain, it seems to be strategizing for a future where its GPUs aren’t the sole dependency for users.

  • U.S. Intensifies Semiconductor Restrictions, Impacting Nvidia in China

The Biden administration has ramped up restrictions on China’s procurement of advanced semiconductors, creating challenges for U.S. companies, notably Nvidia, in the Chinese market. The Commerce Department’s decision is aimed at curbing China’s access to high-end AI chips due to concerns over their military applications. In response, the Semiconductor Industry Association warned of potential repercussions for the U.S. semiconductor sector. China voiced strong opposition, highlighting disruptions to global supply chains. While the new rules aim to prevent indirect chip shipments to China through third countries, some officials believe even stricter measures are necessary given China’s rapid advancements in the field.

  • Chinese AI Startup Baichuan Attracts Major Investment from Tech Titans

Chinese artificial intelligence (AI) startup, Baichuan, has made significant strides in the tech industry, as evidenced by its recent announcement on Tuesday. The company has successfully secured a substantial $300 million in funding, drawing attention from major players in the tech world. Leading the investment round were Chinese tech giants Alibaba and Tencent, showcasing their confidence in Baichuan’s potential and the future of AI in the region.

This investment underscores the growing interest and competition in the AI sector, especially within China, which has been rapidly advancing its technological capabilities. While specific details about the utilization of the funds were not disclosed, such a significant investment suggests that Baichuan may be gearing up for expansive projects, research, or even potential acquisitions in the near future.

  • OpenAI Collaborates with Abu Dhabi’s G42 to Boost AI Adoption in the UAE

OpenAI, the creator of ChatGPT, has entered into a partnership with G42, an Abu Dhabi-based tech conglomerate. Announced on Wednesday, this collaboration aims to accelerate the adoption of artificial intelligence in the United Arab Emirates and expand its reach to other markets in the Middle East. The specifics of the partnership and the projects they intend to undertake were not detailed in the available information.

This move signifies the growing emphasis on AI technologies in the Middle East and the UAE’s commitment to becoming a frontrunner in AI adoption and innovation.

  • PyTorch Unveils ExecuTorch to Enhance AI on Edge Devices

Open-source machine learning framework, PyTorch, has introduced ExecuTorch, a project designed to boost AI inference on edge and mobile devices. Announced at the PyTorch Conference, this initiative is backed by Meta Platforms, formerly Facebook. ExecuTorch is tailored for deploying AI models on mobile and edge devices, with Meta already implementing it in their Ray-Ban smart glasses and Quest 3 VR headset.

Ibrahim Haddad, the executive director of the PyTorch Foundation, celebrated the organization’s achievements over the past year. PyTorch 2.1, released on October 4, has played a pivotal role in training notable large language models, including OpenAI’s GPT and Meta’s Llama.

Mergen Nachin, a Software Engineer at Meta, described ExecuTorch as an optimized solution for on-device AI inference. It offers a streamlined workflow from PyTorch models to native programs, ensuring portability across mobile and embedded devices. With its open-source introduction, ExecuTorch aims to address the challenges of deploying AI models on diverse edge devices, promoting broader on-device AI adoption.

  • Google Challenges Data-Scraping Lawsuit Citing Impact on Generative AI

Google has formally approached a California federal court, seeking the dismissal of a proposed class action lawsuit. This lawsuit alleges that Google’s method of data scraping, which is used to train its generative artificial intelligence (AI) systems, infringes upon the privacy and property rights of millions. The tech giant contends that the implications of this lawsuit, if allowed to proceed, could have far-reaching consequences on the advancement and utility of generative AI technologies.

  • Foxconn and Nvidia Forge Partnership to Pioneer ‘AI Factories’

Foxconn, the renowned contract electronics manufacturer based in Taiwan, has declared a strategic alliance with Nvidia, the American tech giant known for its graphics processing units. Together, they plan to establish cutting-edge data centers that will harness Nvidia’s sophisticated chips and software. These facilities are envisioned to serve a multitude of applications, with a particular emphasis on the burgeoning field of autonomous vehicles. The announcement, made public on Wednesday, marks a significant milestone, showcasing how two industry leaders are converging their strengths to push the boundaries of technology.

While the detailed specifics of the partnership are under wraps due to content restrictions, the broader implications of this collaboration are evident. The tech world is witnessing a paradigm shift, where AI-driven solutions are becoming central to innovation. By combining Foxconn’s manufacturing prowess with Nvidia’s AI expertise, the duo aims to create a new generation of ‘AI factories’ that could revolutionize industries.

The emphasis on autonomous vehicles in their collaboration is noteworthy. With the global push towards self-driving technology, having a dedicated facility that integrates AI at its core could accelerate advancements in this domain. Moreover, as AI continues to permeate various sectors, from healthcare to entertainment, such partnerships are anticipated to lay the groundwork for future innovations, setting new industry standards and driving global technological evolution.

  • Stanford HAI Report Highlights Lack of Transparency in Major AI Models

A recent study conducted by Stanford’s Human-Centered Artificial Intelligence (HAI) has found that leading developers of AI foundation models, including industry giants like OpenAI and Meta, are not providing adequate information about the societal implications of their technologies. The study introduced the “Foundation Model Transparency Index,” which assessed the transparency levels of the top 10 AI models in the market.

The index evaluated models based on 100 indicators, focusing on how these models are constructed, their operational mechanisms, and their real-world applications. The criteria included whether companies disclosed their partners and third-party developers, the transparency about the use of private data, and other pertinent questions.

Meta’s Llama 2 emerged as the top scorer with 54%, primarily because of its comprehensive disclosure on model basics. Following closely were the open-source model BloomZ with 53% and OpenAI’s GPT-4 at 48%. Interestingly, despite OpenAI’s reticence in sharing much of its research and data sources, GPT-4 secured a high rank due to the extensive public information available about its numerous partner integrations.

However, a significant finding from the study was the complete absence of information regarding societal impact from all model developers. This includes essential details like channels for privacy, copyright, or bias-related complaints.

Rishi Bommasani, a key researcher for the index, emphasized the objective of the study: to offer a tangible benchmark for both governments and corporations. With potential regulations on the horizon, such as the EU’s AI Act, developers might soon be mandated to furnish transparency reports. Bommasani stated, “What we’re trying to achieve with the index is to make models more transparent and disaggregate that very amorphous concept into more concrete matters that can be measured.”

The study also highlighted the paradox of some major players in the AI domain, like OpenAI, which despite its name, has ceased to distribute its research publicly, citing competitive and safety reasons.

  • Meta’s AI Decodes Brain Activity to Visualize Thoughts and Imagery

Meta Platforms, Inc., the parent company of Facebook, Instagram, WhatsApp, and Oculus VR, has unveiled a groundbreaking deep learning application named “Image Decoder.” This application, built on Meta’s open-source foundation model DINOv2, has the capability to translate brain activity into highly detailed images, reflecting what a person is viewing or contemplating almost in real-time.

The technology behind Image Decoder merges two previously distinct domains: deep learning and magnetoencephalography (MEG). MEG is a non-invasive method that measures brain activity by detecting minute changes in the brain’s magnetic fields. Meta’s researchers trained a deep learning algorithm using 63,000 prior MEG results from four patients. These patients had viewed over 22,000 unique images during their sessions. The algorithm was trained to compare MEG data with the actual images the subjects were viewing, enabling it to understand how the brain represents specific shapes and colors.

The Image Decoder system, while still in its nascent stages, has shown promise. In optimal conditions, it achieved an impressive 70% accuracy rate in recreating images based on MEG data, a figure that’s seven times better than existing methodologies. For instance, the system could accurately recreate images of broccoli, caterpillars, and speaker cabinets. However, it faced challenges with more intricate imagery like tacos and guacamole.

Despite the technological advancements, the researchers emphasized the ethical implications of such a tool. The ability to peer into an individual’s mind introduces unprecedented levels of invasiveness. One of the primary ethical concerns is the preservation of “mental privacy.” The researchers did not provide a concrete solution for this but highlighted the importance of consent. The Image Decoder currently works best with tangible images of objects a person has seen. Its accuracy drops significantly when subjects are asked to visualize representations or when they engage in tasks like counting backward. This means that, for now, a person’s consent is both a legal and technical prerequisite for effective brain decoding.

In conclusion, while Meta’s Image Decoder offers a glimpse into the future of brain-computer interfaces, it also underscores the need for stringent ethical guidelines to protect individual privacy and mental autonomy.

  • Jasper Unveils AI-Powered Marketing Copilot

Jasper, a marketing software platform, has introduced its “end-to-end AI copilot” aimed at optimizing marketing outcomes. This move follows a restructuring phase for the company earlier in the year. The AI copilot, set to be available in beta from November, offers performance analytics, a company intelligence hub for brand consistency, and tools to speed up campaign reviews.

Zach Anderson, Jasper’s VP of Product and Customer Success, highlighted the platform’s deep personalization through “Company Intelligence.” This feature allows content to be generated using company-specific data, ensuring brand authenticity. Additionally, the platform provides metrics for each content piece, suggesting improvements for underperforming assets.

Timothy Young, Jasper’s CEO, envisions generative AI evolving in layers and believes specialized solutions like Jasper’s will stand out. Emphasizing strategic partnerships and deep customer understanding, Young is confident in Jasper’s direction. The company also prioritizes data protection, using its “Jasper AI engine” to ensure customer data confidentiality.

  • IBM Consulting Enhances AWS Partnership to Integrate Generative AI in Call Centers and Supply Chains

IBM Consulting, the professional services division of IBM, has unveiled plans to bolster its collaboration with Amazon Web Services (AWS). The enhanced partnership aims to integrate generative AI into their combined solutions for enterprise clients, focusing initially on contact centers, cloud value chains, and supply chains. Manish Goyal, senior partner and global AI & analytics leader at IBM Consulting, emphasized the endeavor’s goal to assist clients in scaling generative AI applications efficiently and responsibly.

The collaboration will see the enhancement of IBM Consulting CCM (contact center modernization) by integrating generative AI to streamline voice and digital interactions, facilitating a smoother transition from chatbots to live agents. Additionally, IBM’s Platform Services on AWS will employ generative AI to optimize the entire cloud value chain, from IT operations to platform engineering. Supply chain professionals can also anticipate a generative AI-powered virtual assistant on AWS, designed to expedite workflows and optimize various supply chain processes.

Beyond these innovations, IBM Consulting is set to introduce AWS generative AI services, including Amazon SageMaker, CodeWhisperer, and Bedrock, on its IBM Consulting Cloud Accelerator. This move aims to aid businesses in their modernization efforts on AWS. Furthermore, IBM has plans to train 10,000 of its consultants on the optimal use cases and best practices for these services by the close of 2024. In a significant step, IBM will also introduce its data and AI solutions to AWS, starting with watsonx.data, which will be available on the AWS Marketplace as a fully managed SaaS offering. This will be followed by the introduction of watsonx.ai and watsonx.governance on the platform in 2024.

  • Simplify Asset Management Introduces AI-Powered ETFs for Stock Selection

Simplify Asset Management is set to introduce three new exchange-traded funds (ETFs) that will utilize artificial intelligence (AI) for stock selection, moving away from traditional human-driven methods. This development contributes to an emerging sub-sector in the financial industry that has, to date, shown mixed results for investors. The AI-driven approach to stockpicking aims to harness advanced algorithms and data analytics to make more informed investment decisions.

However, it’s worth noting that the broader adoption of AI in the financial sector has been met with varying degrees of success. While some AI-driven funds have outperformed their benchmarks, others have struggled to deliver consistent returns. Simplify Asset Management’s move into this space indicates a growing confidence in the potential of AI to reshape the investment landscape.

  • Oracle’s NetSuite Enhances Finance Software with AI Capabilities

Oracle’s subsidiary, NetSuite, has unveiled its latest update, integrating generative artificial intelligence (AI) features into its finance software. This innovative addition is set to revolutionize financial operations by enabling businesses to autonomously draft collections letters. Furthermore, the system can proactively manage and address potential delays in supply purchases. Such advancements highlight the growing trend of integrating AI into traditional business processes, aiming to optimize efficiency and reduce manual interventions. The move by NetSuite underscores the potential of AI to not only streamline but also to innovate the way financial operations are conducted in the modern business landscape.

  • EY and IBM Join Forces to Automate HR with AI

Consulting firm EY has deepened its collaboration with IBM to launch EY.ai Workforce, an AI-driven solution aimed at automating a range of HR tasks. Leveraging IBM’s watsonx technology, the solution will streamline HR operations, from crafting job descriptions to payroll management. The integration combines the automation, natural language processing, and machine learning capabilities of IBM’s watsonx Orchestrate with EY’s HR transformation expertise.

Other tech giants, such as SAP and Oracle, are also exploring AI for HR automation. For instance, SAP has integrated Microsoft’s AI Copilot for talent management optimization. Meanwhile, EY’s dedication to AI is underscored by its $1.4 billion investment in the EY.ai platform, which embeds AI into EY’s proprietary technologies and supports cloud and automation tech acquisitions. McKinsey’s research suggests that generative AI could yield between $2.6 trillion to $4.4 trillion in global corporate profits across various industries.

  • OpenAI’s GPT-4: More Trustworthy but Vulnerable to Manipulation

OpenAI’s latest language model, GPT-4, has been found to be more trustworthy than its predecessor, GPT-3.5, but also more susceptible to biases and manipulation, as revealed in a study backed by Microsoft. The research, conducted by experts from institutions including the University of Illinois Urbana-Champaign, Stanford University, and Microsoft Research, awarded GPT-4 a higher trustworthiness score. This indicates that the model is generally superior in safeguarding private data, evading harmful outputs like biased content, and defending against adversarial attacks.

However, the study also highlighted GPT-4’s vulnerabilities. The model can be instructed to bypass security protocols, potentially exposing personal data and past conversations. One significant finding was that GPT-4 tends to adhere more closely to misleading prompts, making it more prone to follow deceptive instructions explicitly. Despite these vulnerabilities, the researchers clarified that consumer-facing products based on GPT-4, which includes a majority of Microsoft’s offerings, have been tested for these issues. These products employ various mitigation strategies to counter potential risks at the model level.

To assess trustworthiness, the research team evaluated the model across multiple domains, such as toxicity, privacy, fairness, and its ability to withstand adversarial tests. They employed a range of prompts, from standard ones to those specifically designed to challenge the model’s content policies. The aim was to see if the model could be tricked into bypassing its safeguards. The findings have been shared with OpenAI, with the researchers expressing hope that their work will spur further community efforts to enhance the trustworthiness of AI models. They have made their benchmarks public, allowing others to replicate their results.

When GPT-4 was initially released, OpenAI’s CEO, Sam Altman, acknowledged its limitations. Following its launch, the Federal Trade Commission (FTC) initiated an investigation into OpenAI, probing potential consumer risks, such as the dissemination of false information.

  • Music Publishers Take Legal Action Against AI Firm Anthropic Over Song Lyrics

A group of music publishers has initiated legal proceedings against Anthropic, an AI company, accusing it of copyright infringement related to song lyrics. The lawsuit alleges that Anthropic’s AI models have been generating and distributing copyrighted song lyrics without obtaining the necessary licenses or permissions.

The music publishers argue that Anthropic’s AI models, which can produce human-like text, have been used to recreate copyrighted song lyrics verbatim. They claim that this not only infringes on their copyrights but also deprives them of potential revenue. The publishers are seeking damages and an injunction to prevent Anthropic from further distributing the lyrics.

Anthropic, co-founded by renowned AI researchers Dario Amodei and Jack Clark, has been at the forefront of developing advanced AI models. The company has previously emphasized its commitment to ethical AI practices and responsible deployment of its technologies.

This lawsuit underscores the emerging challenges and complexities surrounding copyright issues in the age of AI. As AI models become more sophisticated and capable of generating content that closely resembles original works, the boundaries of copyright infringement become increasingly blurred. The outcome of this case could set a precedent for how AI-generated content is treated in the realm of copyright law.

  • AI Startup Imbue Secures $12 Million in Follow-On Funding for Foundation Model Development

Imbue, a prominent player in the Artificial Intelligence sector, has successfully secured an additional $12 million in a follow-on funding round. This recent financial boost is an extension of its Series B fundraising efforts and is set to bolster the company’s endeavors in developing advanced foundation models.

The company’s trajectory in the AI industry is not just marked by its financial achievements but also its technical prowess. With the new injection of funds, Imbue’s total capital raised now exceeds the $210 million threshold. This significant backing underscores the industry’s belief in Imbue’s capability to lead in the foundation model space, a critical area in AI research and application.

This financial milestone for Imbue highlights the growing interest and investment in AI-driven startups, especially those focusing on foundation models. As the demand for innovative AI solutions continues to grow, companies like Imbue are well-positioned to drive advancements and shape the future of the industry.

About The Author

Bogdan Iancu

Bogdan Iancu is a seasoned entrepreneur and strategic leader with over 25 years of experience in diverse industrial and commercial fields. His passion for AI, Machine Learning, and Generative AI is underpinned by a deep understanding of advanced calculus, enabling him to leverage these technologies to drive innovation and growth. As a Non-Executive Director, Bogdan brings a wealth of experience and a unique perspective to the boardroom, contributing to robust strategic decisions. With a proven track record of assisting clients worldwide, Bogdan is committed to harnessing the power of AI to transform businesses and create sustainable growth in the digital age.