3.8.2023
In this month's edition, AI advancements include OpenAI's newsroom collaboration with AP, Meta's innovative text-to-image tool CM3leon, industry giants' safety pledges for AI, Google's medical tool Med-PaLM 2's trials, and global discussions on AI's regulation and the rise of Superintelligence.
Cedric May
Chief Technology Officer
Artificial Intelligence (AI) has made impressive strides in recent years. Notable advancements were announced in July 2023, including how AI intertwines with the medical sector, forges business partnerships, impacts the news industry, and shapes the future of computer graphics.
For regular updates, don't forget to subscribe to our weekly newsletter on LinkedIn. Enjoy the read, and let's dive right in to the fascinating world of AI.
In this edition, you'll discover:
Google’s Med-PaLM 2, an AI tool for answering medical questions, is being tested at the Mayo Clinic research hospital and other locations since April, according to a report by The Wall Street Journal. Med-PaLM 2 is an updated version of PaLM 2, which was announced at Google I/O in May 2023, and is believed to be more useful for healthcare conversations than other chatbots. While there are some inaccuracies, Med-PaLM 2 performs well in most metrics, and customers testing the tool will have control over their data. According to Google senior research director Greg Corrado, Med-PaLM 2 is still in its early stages, but has the potential to expand the places where AI can be beneficial in healthcare by 10-fold.
KPMG has announced a confident plan to invest a staggering $2 billion in artificial intelligence (AI) and cloud services in partnership with Microsoft over the next five years. By doing so, the partnership is expected to generate over $12 billion in revenue. This significant investment will allow KPMG to further automate aspects of its tax, audit, and consulting services. The company will be able to provide even faster analysis, spend more time on strategic advice, and help more companies integrate AI into their operations. A considerable portion of the investment will be allocated towards generative AI, which can cut costs and yield new efficiencies. By utilizing this technology, KPMG will strengthen its work related to environmental, social, and governance issues. Additionally, KPMG will have early access to an AI assistant called Microsoft 365 Copilot. It is worth noting that other Big Four firms have also invested billions of dollars in cloud-based and other technologies to strengthen their audit, consulting, and tax businesses with Microsoft's AI offerings.
The Associated Press made a two-year deal with OpenAI, parent company to ChatGPT, to share select news content and technology. The deal marks one of the first official news-sharing agreements between a major US news company and an AI firm. OpenAI will license some of the AP's text archive dating back to 1985 to help train its AI algorithms. In exchange, the AP will get access to OpenAI's technology and product expertise. The two firms are still working through the technical details of how the sharing will work on the back end. The partnership with OpenAI is meant to help the AP understand responsible use cases for generative AI in news products and services in the future. The news industry is grappling with ways to best leverage AI to improve output. The AP hopes to be an industry leader in developing standards and best practices around generative AI for other newsrooms.
Playground has successfully raised $40.8M to drive innovation in computer graphics and create intelligent models. Their confident plan is to commence with image processing, aiming to achieve state-of-the-art capabilities and empower individuals to create beyond their previous limitations. The plan encompasses developing a modern graphics editor, advancing research in computer vision, constructing multi-task models, and enabling an exponential increase in the number of people capable of creating.
Interest and research in generative AI models has grown recently due to advancements in natural language processing and image generation. Meta is demonstrating CM3leon, a single foundation model that can generate both text and images.
CM3leon is the first multimodal model trained with a recipe that adapts from text-only language models. This recipe is simple and produces a strong model. CM3leon can generate sequences of text and images conditioned on arbitrary sequences of other image and text content. This expands previous models that were either only text-to-image or only image-to-text.
CM3leon is a large-scale multitask model that improves image and text generation performance. It performs well on image caption generation, visual question answering, text-based editing, and conditional image generation. Compared to Google's Parti, CM3leon achieves an FID score of 4.88 on the zero-shot MS-COCO benchmark, making it the new state of the art in text-to-image generation. CM3leon also excels at generating complex compositional objects and performs well in vision-language tasks such as visual question answering and long-form captioning.
CM3leon is capable of text-guided image generation and editing, text-to-image, text tasks, structure-guided image editing, object-to-image, and segmentation-to-image. Its decoder-only transformer architecture is trained using a licensed dataset.
As generative AI models like CM3leon become more sophisticated, it's important to be transparent about any biases in the training data. By doing so, we can create more accurate and equitable models for everyone.
China has unveiled provisional rules to govern generative AI services for China-based users. The rules aim to balance development and security, requiring adherence to core socialist values and prohibiting everything from pornography and terrorism to racism and content that threatens China's national security. Algorithms that can influence public opinions must be registered, and providers must obtain an administrative license. Algorithms must not be discriminatory based on factors such as ethnicity, gender, age, occupation, or health. Service providers must identify and stop the generative process for illegal content, and regulators have the right to know the specifics of a generative AI model. Finally, China is calling for self-reliant innovation in AI while encouraging international cooperation.
Several large generative AI companies based in the US will mark their content with watermarks. Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI have agreed to eight voluntary commitments related to the oversight and use of generative AI, including watermarking, according to a fact sheet from the White House published on July 21. The agreement follows a White House statement in March about concerns regarding the misuse of AI. It also comes at a time when regulators are establishing procedures for managing the effects of generative AI on technology and people since ChatGPT put AI content in the public eye in November 2022.
The eight AI safety commitments include:
The watermark commitment involves developing a way to mark text, audio or visual content as machine-generated. It will apply to any publicly available generative AI content created after the watermarking system is locked in. Former Microsoft Azure global vice president and current Cognite chief product officer Moe Tanabian supports government regulation of generative AI, as there are opportunities for malicious actors to take advantage of generative AI. He compared the current era of generative AI with the rise of social media, including possible downsides like the Cambridge Analytica data privacy scandal and other misinformation during the 2016 election.
Renowned billionaire financier George Soros confidently warns of the grave danger that artificial intelligence poses to democracies and its potential to become a powerful tool for authoritarians. Soros acknowledges that A.I. has the potential to enable closed societies to monitor their citizens. He further highlights that global regulations for A.I. are unattainable due to the conflict between open and closed societies. Although Soros is instinctively opposed to A.I., he acknowledges that he does not know how to stop it. Despite experts calling for the regulation of A.I., Soros firmly believes that regulations must be globally enforceable. In the coming months, Soros’ Open Society Foundations will cut at least 40% of its staff. Soros believes that A.I. will play a pivotal role in the next year's U.S. elections and, if left unchecked, will undoubtedly pose a significant threat.
Superintelligence is a powerful technology that could solve major global issues, but it also poses a high risk of disempowering or even destroying humanity. OpenAI is taking proactive measures to prepare for its arrival within a decade, including developing new institutions for governance and tackling the problem of superintelligence alignment. They are dedicating 20% of their computing power over the next four years to solve this problem, assembling a team of top researchers and engineers to work on it. Their work complements existing efforts at OpenAI to improve AI safety.
Anthropic's AI chatbot, Claude 2, is now available for public testing in the US and UK. Claude 2 has a conversational tone and can perform tasks like creating summaries, coding, and translation. It is guided by a set of principles called a "constitution" to revise its responses by itself. Claude 2 is an improvement upon the previous model, with better math, coding, and reasoning skills. Additionally, it has a larger context window of around 75,000 words, and is less likely to produce harmful content. Despite being limited to data up to December 2022, it can still answer queries about recently published web pages.
Microsoft’s AI chatbot, Bing Chat, will now be available on non-Microsoft browsers, including Google Chrome and Apple’s Safari. This will give users outside of Microsoft products, such as the Bing mobile app and Microsoft Edge browser, access to the ChatGPT-like AI chatbot. Microsoft confirmed the expansion but did not announce which other browsers will be supported yet. Bing Chat is powered by OpenAI’s GPT-4 model, and it seems to be testing a native dark theme. However, the experience in other browsers has some limitations, including only supporting five messages per conversation in Chrome, instead of the 30 available in Microsoft Edge, and limiting the character count to 2,000, instead of the 3,000 supported by Edge. Bing Chat has been integrated with Skype and has even launched a version for enterprise use, which includes business-focused data privacy and governance controls.
Thank you for joining us for July's edition of FrontNow - Trends in AI! During this edition, we explored the diverse landscape of artificial intelligence, from Google's cutting-edge work in healthcare AI to China's regulatory efforts and George Soros' concerns about the societal impact of AI. As we look ahead to August, let's continue to embrace disruption. We invite you to join us again next month as we continue to explore world of AI.
Artificial Intelligence (AI) has made impressive strides in recent years. Notable advancements were announced in July 2023, including how AI intertwines with the medical sector, forges business partnerships, impacts the news industry, and shapes the future of computer graphics.
For regular updates, don't forget to subscribe to our weekly newsletter on LinkedIn. Enjoy the read, and let's dive right in to the fascinating world of AI.
In this edition, you'll discover:
Google’s Med-PaLM 2, an AI tool for answering medical questions, is being tested at the Mayo Clinic research hospital and other locations since April, according to a report by The Wall Street Journal. Med-PaLM 2 is an updated version of PaLM 2, which was announced at Google I/O in May 2023, and is believed to be more useful for healthcare conversations than other chatbots. While there are some inaccuracies, Med-PaLM 2 performs well in most metrics, and customers testing the tool will have control over their data. According to Google senior research director Greg Corrado, Med-PaLM 2 is still in its early stages, but has the potential to expand the places where AI can be beneficial in healthcare by 10-fold.
KPMG has announced a confident plan to invest a staggering $2 billion in artificial intelligence (AI) and cloud services in partnership with Microsoft over the next five years. By doing so, the partnership is expected to generate over $12 billion in revenue. This significant investment will allow KPMG to further automate aspects of its tax, audit, and consulting services. The company will be able to provide even faster analysis, spend more time on strategic advice, and help more companies integrate AI into their operations. A considerable portion of the investment will be allocated towards generative AI, which can cut costs and yield new efficiencies. By utilizing this technology, KPMG will strengthen its work related to environmental, social, and governance issues. Additionally, KPMG will have early access to an AI assistant called Microsoft 365 Copilot. It is worth noting that other Big Four firms have also invested billions of dollars in cloud-based and other technologies to strengthen their audit, consulting, and tax businesses with Microsoft's AI offerings.
The Associated Press made a two-year deal with OpenAI, parent company to ChatGPT, to share select news content and technology. The deal marks one of the first official news-sharing agreements between a major US news company and an AI firm. OpenAI will license some of the AP's text archive dating back to 1985 to help train its AI algorithms. In exchange, the AP will get access to OpenAI's technology and product expertise. The two firms are still working through the technical details of how the sharing will work on the back end. The partnership with OpenAI is meant to help the AP understand responsible use cases for generative AI in news products and services in the future. The news industry is grappling with ways to best leverage AI to improve output. The AP hopes to be an industry leader in developing standards and best practices around generative AI for other newsrooms.
Playground has successfully raised $40.8M to drive innovation in computer graphics and create intelligent models. Their confident plan is to commence with image processing, aiming to achieve state-of-the-art capabilities and empower individuals to create beyond their previous limitations. The plan encompasses developing a modern graphics editor, advancing research in computer vision, constructing multi-task models, and enabling an exponential increase in the number of people capable of creating.
Interest and research in generative AI models has grown recently due to advancements in natural language processing and image generation. Meta is demonstrating CM3leon, a single foundation model that can generate both text and images.
CM3leon is the first multimodal model trained with a recipe that adapts from text-only language models. This recipe is simple and produces a strong model. CM3leon can generate sequences of text and images conditioned on arbitrary sequences of other image and text content. This expands previous models that were either only text-to-image or only image-to-text.
CM3leon is a large-scale multitask model that improves image and text generation performance. It performs well on image caption generation, visual question answering, text-based editing, and conditional image generation. Compared to Google's Parti, CM3leon achieves an FID score of 4.88 on the zero-shot MS-COCO benchmark, making it the new state of the art in text-to-image generation. CM3leon also excels at generating complex compositional objects and performs well in vision-language tasks such as visual question answering and long-form captioning.
CM3leon is capable of text-guided image generation and editing, text-to-image, text tasks, structure-guided image editing, object-to-image, and segmentation-to-image. Its decoder-only transformer architecture is trained using a licensed dataset.
As generative AI models like CM3leon become more sophisticated, it's important to be transparent about any biases in the training data. By doing so, we can create more accurate and equitable models for everyone.
China has unveiled provisional rules to govern generative AI services for China-based users. The rules aim to balance development and security, requiring adherence to core socialist values and prohibiting everything from pornography and terrorism to racism and content that threatens China's national security. Algorithms that can influence public opinions must be registered, and providers must obtain an administrative license. Algorithms must not be discriminatory based on factors such as ethnicity, gender, age, occupation, or health. Service providers must identify and stop the generative process for illegal content, and regulators have the right to know the specifics of a generative AI model. Finally, China is calling for self-reliant innovation in AI while encouraging international cooperation.
Several large generative AI companies based in the US will mark their content with watermarks. Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI have agreed to eight voluntary commitments related to the oversight and use of generative AI, including watermarking, according to a fact sheet from the White House published on July 21. The agreement follows a White House statement in March about concerns regarding the misuse of AI. It also comes at a time when regulators are establishing procedures for managing the effects of generative AI on technology and people since ChatGPT put AI content in the public eye in November 2022.
The eight AI safety commitments include:
The watermark commitment involves developing a way to mark text, audio or visual content as machine-generated. It will apply to any publicly available generative AI content created after the watermarking system is locked in. Former Microsoft Azure global vice president and current Cognite chief product officer Moe Tanabian supports government regulation of generative AI, as there are opportunities for malicious actors to take advantage of generative AI. He compared the current era of generative AI with the rise of social media, including possible downsides like the Cambridge Analytica data privacy scandal and other misinformation during the 2016 election.
Renowned billionaire financier George Soros confidently warns of the grave danger that artificial intelligence poses to democracies and its potential to become a powerful tool for authoritarians. Soros acknowledges that A.I. has the potential to enable closed societies to monitor their citizens. He further highlights that global regulations for A.I. are unattainable due to the conflict between open and closed societies. Although Soros is instinctively opposed to A.I., he acknowledges that he does not know how to stop it. Despite experts calling for the regulation of A.I., Soros firmly believes that regulations must be globally enforceable. In the coming months, Soros’ Open Society Foundations will cut at least 40% of its staff. Soros believes that A.I. will play a pivotal role in the next year's U.S. elections and, if left unchecked, will undoubtedly pose a significant threat.
Superintelligence is a powerful technology that could solve major global issues, but it also poses a high risk of disempowering or even destroying humanity. OpenAI is taking proactive measures to prepare for its arrival within a decade, including developing new institutions for governance and tackling the problem of superintelligence alignment. They are dedicating 20% of their computing power over the next four years to solve this problem, assembling a team of top researchers and engineers to work on it. Their work complements existing efforts at OpenAI to improve AI safety.
Anthropic's AI chatbot, Claude 2, is now available for public testing in the US and UK. Claude 2 has a conversational tone and can perform tasks like creating summaries, coding, and translation. It is guided by a set of principles called a "constitution" to revise its responses by itself. Claude 2 is an improvement upon the previous model, with better math, coding, and reasoning skills. Additionally, it has a larger context window of around 75,000 words, and is less likely to produce harmful content. Despite being limited to data up to December 2022, it can still answer queries about recently published web pages.
Microsoft’s AI chatbot, Bing Chat, will now be available on non-Microsoft browsers, including Google Chrome and Apple’s Safari. This will give users outside of Microsoft products, such as the Bing mobile app and Microsoft Edge browser, access to the ChatGPT-like AI chatbot. Microsoft confirmed the expansion but did not announce which other browsers will be supported yet. Bing Chat is powered by OpenAI’s GPT-4 model, and it seems to be testing a native dark theme. However, the experience in other browsers has some limitations, including only supporting five messages per conversation in Chrome, instead of the 30 available in Microsoft Edge, and limiting the character count to 2,000, instead of the 3,000 supported by Edge. Bing Chat has been integrated with Skype and has even launched a version for enterprise use, which includes business-focused data privacy and governance controls.
Thank you for joining us for July's edition of FrontNow - Trends in AI! During this edition, we explored the diverse landscape of artificial intelligence, from Google's cutting-edge work in healthcare AI to China's regulatory efforts and George Soros' concerns about the societal impact of AI. As we look ahead to August, let's continue to embrace disruption. We invite you to join us again next month as we continue to explore world of AI.
Read next
Stay Updated With Frontnow's News & Cutting-Edge AI Innovation.
Your partner in redefining e-commerce, leveraging AI for hyper-personalized online experiences and superior product discovery.
