Weekly AI - 5 Nov 2023
Hello readers! Welcome to this week's edition of the AI News Roundup. I'm excited to share some of the latest happenings in the world of artificial intelligence. From major investments in AI startups to groundbreaking new capabilities in generative models, there are fascinating developments across the AI landscape.
In this issue, we'll look at how AI is transforming areas like drug discovery, robotics, creative tools, and more. Key players like Google, Microsoft, Anthropic and others are pushing boundaries with big funding and new releases. We'll also examine some of the growing concerns around responsible and ethical AI development.
My goal is to keep you informed on how AI is evolving rapidly and changing the world around us. I'll do my best to break down complex topics into digestible snippets. As always, I'm happy to hear your thoughts and feedback. Now let's jump in!
This newsletter is also available in Spanish and Catalan.
Biden Signs Executive Order to Develop AI Safety Guidelines
President Biden signed an executive order directing government agencies to develop guidelines and standards for AI safety and preventing discrimination. The order tasks agencies like the National Institute of Standards and Technology (NIST) and Department of Homeland Security (DHS) with creating best practices around content authentication, AI security, and testing models before public release. It also aims to protect privacy, support workers impacted by AI, and advance US leadership in AI. While not a permanent law, the order signals the administration is prioritizing oversight of generative AI like chatbots.
ChatGPT Gets Major Upgrade With GPT-4 All Tools
ChatGPT's new "All Tools" upgrade gives users access to the full range of GPT-4 capabilities in one place for the first time. Instead of switching between different modes, the AI assistant can now seamlessly handle text queries, generate images with DALL-E 3, analyze data, and browse the web. This allows for more integrated workflows without needing third-party plugins. The consolidated experience shows ChatGPT evolving beyond a pure text-based chatbot. Users can now upload documents like PDFs for analysis directly in the tool. With image generation built-in, new creative applications are possible by pairing text prompts with AI-created visuals. The upgraded ChatGPT pushes the boundaries of what's possible with a single AI assistant.
The Beatles Release Final Song "Now and Then" Using AI to Revive John Lennon's Voice
The remaining Beatles members Paul McCartney and Ringo Starr have released a new song called "Now and Then" using AI technology to recreate the voice of John Lennon. The track was originally recorded before Lennon's death in 1980 but was not released due to objections from George Harrison. Decades later, McCartney has used new AI capabilities to isolate and clean up Lennon's vocals from the original tape, finally completing the song. With this release, McCartney finally accomplishes his decades-long goal of completing the unfinished song from his late bandmate, signifying the official conclusion of new music from the iconic band The Beatles. However, McCartney acknowledges AI music technology is a double-edged sword, marveling at its capabilities while admitting it gives him pause about potential misuse.
In a lower key note, Johnny Cash's iconic voice was cloned to sing Taylor Swift's "Blank Space" in an unsettling AI experiment, raising concerns about technology replacing human creativity.
Brave's Leo AI Assistant Now Available to All Desktop Users
Brave released its AI assistant Leo to all desktop users after testing since August. Leo is based on Llama 2 and Anthropic's Claude LLMs and can summarize webpages, translate text, and rewrite phrases. It is free, but a $15/month premium version offers faster, more accurate responses. Brave says Leo's requests are anonymous and responses are not stored. This release comes as browsers compete to offer built-in AI features, and Brave looks for revenue streams like its own search API.
Google's AI-Powered Search to Include Ads Soon
Google announced that it will start experimenting with new ad formats tailored for its AI-powered search called Search Generative Experience (SGE). During an earnings call, CEO Sundar Pichai said SGE is a "long-term play" and ads need to "transition well." The integration of ads into SGE will be key for Google's core search advertising business, which generated $44 billion last quarter. While SGE is currently opt-in, Pichai envisions it becoming "just how search works." The company acknowledges the importance of "advertisers still having the opportunity to reach potential customers" as SGE rolls out more widely.
Google Bets Big on AI with a $2 Billion Investment in Anthropic
Google has agreed to invest up to $2 billion in Anthropic, the AI startup behind Claude chatbot that competes with OpenAI's popular ChatGPT. The deal includes $500 million upfront and $1.5 billion over time, valuing Anthropic at over $4 billion. This huge investment signals Google's confidence in Anthropic's approach to developing responsible and self-aware AI. It also intensifies the AI rivalry between tech giants like Google, Microsoft and others who are racing to lead in this transformative technology.
Jina AI Launches Cutting-Edge 8K Text Embedding Model
Berlin-based Jina AI unveiled its latest text embedding model, jina-embeddings-v2. This open-source model supports an impressive 8K context length, putting it on par with proprietary offerings like OpenAI's text-embedding-ada-002. According to Jina AI's CEO Dr. Han Xiao, this marks a milestone in democratizing AI by making such a high-performing model freely available. He stated that jina-embeddings-v2 empowers the community with tools once limited to private ecosystems. The launch represents a leap towards Jina's vision of advancing AI through open source and open science.
Boston Dynamics' Spot the Robot Dog Becomes an AI-Powered Tour Guide
Boston Dynamics shared details last week on how they used AI models like ChatGPT to give their Spot robot dog basic conversational skills. By combining a large language model, vision systems, speech recognition and text-to-speech, their team was able to create a “tour guide” demo where Spot could describe what it saw and answer questions. While limited, the project shows how far robotics has come in integrating AI to enable more natural interaction. As the company put it, “a world in which robots can generally understand what you say and turn that into useful action is probably not that far off.” This blend of robotics and AI could someday allow robots to communicate effortlessly alongside people. Don't miss the video here.
Quora's Poe chatbot app introduces monetization for AI creators
The AI chatbot app Poe from question-and-answer platform Quora launched a new program to pay bot creators for their efforts. Creators will earn a share of subscription revenue from users who subscribe to Poe after engaging with their bots. This is one of the first monetization programs for AI chatbot creators. Quora CEO Adam D'Angelo said this will enable smaller AI groups and researchers to reach the public. The program comes as Poe aims to compete in the growing AI chatbot app market.
AlphaFold's Latest Model Advances Drug Discovery
DeepMind released a new version of its AI system AlphaFold, which can predict the structures of proteins and other molecules. According to DeepMind, this latest AlphaFold model is more useful for drug discovery because it can predict how proteins interact with ligands - molecules that bind to proteins and affect cell communication. This could help scientists identify and design new drug compounds. DeepMind claims the new AlphaFold drastically outperforms previous versions on drug discovery-relevant prediction problems like antibody binding. While not perfect yet, especially for RNA prediction, this update shows AI's potential to enhance understanding of human biology for medicine.
LinkedIn Rolls Out AI-Powered Reading and Writing Tools for 1 Billion Users
Professional networking platform LinkedIn announced new AI features for its 1 billion users. The company is leveraging OpenAI's GPT technology to provide personalized reading digests and writing suggestions. The goal is to encourage more engagement on the platform. LinkedIn is starting with its paid Premium subscribers but may expand the features more broadly if they prove valuable. It's the latest move by LinkedIn to stay relevant and capitalize on interest in generative AI.
Google Launches AI-Generated Product Imagery for US Advertisers
Google announced the launch of new generative AI tools named Product Studio that allow US merchants and advertisers to create free product imagery simply by typing text prompts. The AI-powered tools can modify existing product photos by changing colors or removing backgrounds, or generate completely new product images like a skincare item surrounded by seasonal scenery. Google says this will help businesses reuse assets across campaigns without costly reshoots. The new imagery capabilities are now available to Merchant Center Next users and the Google and YouTube app on Shopify merchants in the US.
HubSpot Acquires AI Startup Clearbit to Enhance its Platform
HubSpot announced its acquisition of Clearbit, an AI-powered B2B data provider. Clearbit's technology uses large language models to enrich customer data with real-time external information. According to HubSpot CEO Yamini Rangan, combining Clearbit's data with HubSpot's AI platform will create a "powerful, winning combination" for customers. She said that gathering company data has gotten easier but analyzing and using it remains challenging. The acquisition aims to help HubSpot customers enrich their internal data with more context.
Stability AI Launches AI-Powered 3D Model Generator
AI startup Stability AI unveiled its latest creation - Stable 3D, an AI system that generates 3D models and textures from text descriptions. The company claims Stable 3D will allow non-expert users like graphic designers and game developers to quickly create high-quality 3D assets, enabling new creative possibilities. However, some experts note Stability has been accused of training their AI models on copyrighted data without proper licensing, raising concerns about how 3D models generated by Stable 3D could potentially infringe on others' IP rights.
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!