Weekly AI - 26 Nov 2023
Hello readers! Welcome to Weekly AI. In this issue, we'll learn about major leadership changes at OpenAI, tools that allow anyone to generate code from sketches, how the use of AI can help detect Alzheimer's, and breakthroughs in weather forecasting from DeepMind.
I always appreciate your thoughts and questions. Please keep sending feedback so we can have an open dialogue about artificial intelligence and its impacts. For now, let's jump in!
This newsletter is also available in Spanish and Catalan.
Dramatic Leadership Changes at OpenAI Raise Questions About AI Safety
OpenAI, the nonprofit AI lab behind chatbot ChatGPT, has undergone major leadership changes in recent days. CEO Sam Altman was abruptly fired by the board of directors on November 17th, with president Greg Brockman resigning immediately after. As described in this timeline, the shakeup shocked the tech world and led the majority of OpenAI's staff to threaten resignation.
Many observers believe safety concerns about OpenAI's powerful AI models played a key role. As Tomas Pueyo explains in his two posts (link1 and link2), OpenAI was founded specifically to ensure advanced AI benefits humanity. However, there were reported tensions between Altman and the board over the pace of commercialization versus investigating potential risks.
After a turbulent few days, Altman and the board reached an agreement for him to return as CEO. But the drama has fueled debate about balancing rapid innovation in AI with containment of potentially dangerous capabilities. As AI progresses towards human-level intelligence, difficult questions around ethics, priorities, and control remain unresolved.
Bard Can Now Answer Questions About YouTube Videos
Google has announced an update to its AI chatbot Bard that allows it to understand and answer specific questions about YouTube videos. For example, if you ask the chatbot how many eggs are used in a recipe shown in a video, it can analyze the footage and provide an answer. This new capability builds on Bard's existing YouTube extension that was launched in September. While Bard could previously search YouTube and recommend videos, it can now have a "richer conversation" about the videos themselves.
Claude 2.1 Launches with 200K Context Window and Reduced Hallucination Rates
AI company Anthropic released Claude 2.1, their latest conversational AI model. Claude 2.1 doubles the amount of context it can process to 200,000 tokens, allowing it to summarize long, complex documents like financial reports and literature. It also reduces false statements by 2x compared to the previous version. According to Anthropic, Claude 2.1 shows significant improvements in comprehension and honesty.
Tldraw and GPT-4V allow anyone to sketch a website UI and generate working code
The tldraw team released makereal.tldraw.com, allowing anyone with an OpenAI API key to sketch a website user interface on their infinite canvas and use GPT-4V to generate functioning HTML/CSS code from the drawing. As described in their recent blog post, this tool enables rapid visualization and iteration of website prototypes directly on the canvas through conversations with the AI. The tldraw team believes this technology barely scratches the surface of what’s possible and hopes users will build their own creative experiments.
DeepMind Unveils Lyria - Its Most Advanced AI Music Generator
Google's DeepMind division announced Lyria, its most advanced AI system for generating high-quality, original music. Lyria can create instrumental tracks, vocals, and complete songs in different musical styles. DeepMind is launching two experiments with Lyria - Dream Track on YouTube to generate soundtracks sung by famous artists, and Music AI tools for artists and producers to aid creativity. These launches represent a milestone in AI's ability to produce compelling, creative music. DeepMind promises responsible development of the technologies, involving musicians and putting safeguards like watermarking in place.
Bard opens to teens with safety features and learning tools
Google announced the expansion of its AI chatbot Bard to teenagers in most countries. This is an important step as Bard can help teens learn new skills and find inspiration. However, Google implemented safety features and content policies to prioritize teen safety, such as training Bard to avoid inappropriate content.
Common Sense Media Rates Popular AI Tools - Many Flagged as Unsafe for Kids
An independent non-profit organization, Common Sense Media, has published safety and ethics ratings for 10 popular AI apps and services. In the reviews published on November, the organization flagged several widely-used AI generative models, including DALL-E, Stable Diffusion and Snapchat's My AI, as potentially unsafe for use by children. Common Sense Media raised concerns about inappropriate content, reinforcement of unfair biases, privacy issues and spreading of misinformation by these AI systems.
Meta Taking Steps to Ensure Transparency of AI-Generated Content in Political Ads
Meta announced new rules requiring political advertisers to disclose when ads contain AI-generated or altered images, videos, or audio. Starting next year, ads with realistic but fake depictions of people or events must be labeled to inform users. Meta says this policy aims to prevent the spread of misinformation, though some critics argue it does not go far enough. The move comes as lawmakers push for broader regulations around AI use in campaign materials ahead of the 2024 elections.
YouTube outlines approach to responsible AI innovation, including new disclosure requirements
YouTube recently announced they will roll out updates over the coming months to inform viewers when content they see contains realistic altered or synthetic material, including that made with AI tools. Specifically, creators will be required to disclose if they uploaded content that is altered or synthetic, including AI-generated content, that depicts realistic events or actions. YouTube says labels will be added and content may be removed if it violates policies. They state this will help address risks of misinformation and clarify the nature of new forms of synthetic storytelling enabled by AI for viewers.
ChatGPT and Other AI Now Have Access to 20,000+ Zapier Actions
AI Actions announced this week that ChatGPT and other AI platforms can now integrate with and trigger any of the 20,000+ actions from the popular automation platform Zapier. This allows AI systems to perform tasks like sending Slack messages, creating HubSpot leads, and more. Developers must enable access to specific Zapier actions through the AI Actions platform. According to AI Actions, this integration removes previous hurdles like authentication and API setup for AI platforms.
AI Brain Scan Analysis Shows Promise for Detecting Alzheimer's
A new machine learning approach that analyzes brain scans was able to detect Alzheimer's disease with over 90% accuracy in a study published on November. This AI analysis of PET scans could be a breakthrough for both clinicians trying to diagnose Alzheimer's early and scientists working on new treatments. As lead author Dr. Lu states, "An accurate and convenient diagnostic aid like this could revolutionize how we detect Alzheimer's and track disease progression." The ability to catch Alzheimer's earlier and monitor it more precisely gives doctors a better shot at slowing its development. And for researchers, identifying people accurately at different disease stages is key to testing whether new therapies targeted at specific brain changes actually work.
Machine learning model forecasts weather up to 10 days with record accuracy
Researchers at Google DeepMind have developed a machine learning model called GraphCast that can forecast global weather conditions up to 10 days in advance with greater accuracy than existing traditional weather prediction systems. According to the results published on November in the journal Science, GraphCast outperformed the European Centre for Medium-Range Weather Forecasts' (ECMWF) high-resolution forecast system on 90% of tested metrics over the 10-day predictions.
Resources
Check this list of GPTs, to see what they capable of
Common Sense Media. Check this site if you have children and want to make sure that what your kids are watching is adequate for their age.
That wraps up this week's AI news! If you found this informative, don't forget to share it with friends and colleagues. And be sure to subscribe to get next week's news straight to your inbox. Thanks for reading!