GSV's AI News & Updates (08/29/23)

Databricks $500M Series I, Stable Audio Music Generator, Consumer GenAI Usage Patterns, Ello, Delphi, Learn.xyz, MagicSchool, Google's Responsible AI Fund

You are receiving this weekly update on the intersection of AI and education, as part of our GSV community – we hope this will be a helpful resource as we navigate this evolving and exciting intersection. Please feel free to send over any relevant resources you may come across (claire@gsv.com) – we’d love to learn and hear from you!

---

General 🚀

• Time magazine announces TIME100 Most Influential People in AI: across leaders, innovators, shapers and thinkers. This group of 100 individuals is in many ways a map of the relationships and power centers driving the development of AI. They are rivals and regulators, scientists and artists, advocates and executives—the competing and cooperating humans whose insights, desires, and flaws will shape the direction of an increasingly influential technology.

  

• Stability AI launches Stable Audio, an AI-powered music generator: today marks the release of Stable Audio, a tool that Stability claims is the first capable of creating “high-quality,” 44.1 kHz music for commercial use via a technique called latent diffusion. Trained on audio metadata as well as audio files’ durations, Stability says that Audio Diffusion’s underlying, roughly 1.2B parameter model affords greater control over the content and length of synthesized audio.

• Adobe’s Firefly generative AI models are now generally available: the company is going to use what it calls “generative credits” to measure how often users interact with these models. Every time you click “generate” to create a Firefly image, you’ll consume one credit

  

  .

• Anthropic releases Claude Pro, joining the paid chatbot field: subscribers can now gain 5x more usage of their latest model, Claude 2, and priority access during high-traffic periods.

• Adept open-sources Persimmon-8B: the most powerful fully permissively-licensed language model with <10 billion parameters. Adept is working towards an AI agent that can help people do anything they need to do on a computer.

• Apple Boosts Spending to Develop Conversational AI: Apple's budget for training AI is now millions of dollars per day.

• Google Gives Public Data Project An AI Makeover: Google has launched an AI-driven interface for Data Commons, simplifying data access and analysis through natural language queries.

• Amazon debuts generative AI tools that helps sellers write product descriptions: though Amazon doesn’t specifically say, it seems that the retailer likely scoured its own listing data to train its machine learning models. “Our models learn to infer product information through the diverse sources of information, latent knowledge, and logical reasoning that they learn”.

• Salesforce embeds conversational AI across the platform with Einstein Copilot: The idea is to let users ask the bot in a conversational way to get information that would often take several clicks and knowledge of how to do it. A salesperson could research new accounts, a product manager could create a customized storefront for a new product launch.

• Your Gmail and Instagram are training AI. There’s little you can do about it. It’s your Gmail. It’s also Google’s artificial intelligence factory. Unless you turn it off, Google uses your Gmail to train an AI to finish other people’s sentences. You might be accustomed to them selling your data or using it to target you with ads. But now they’re using it to create lucrative new technologies that could upend the economy — and make Big Tech even bigger.

• How Perplexity.ai Is Pioneering The Future Of Search: Perplexity.ai is an “answer engine” that places users, not advertisers, at its center. This shift promises to transform how we discover, access, and consume knowledge online—and, with that, the structure of the internet as we know it today.

• How Are Consumers Using Generative AI?: in which categories do incumbents dominate—and where are new companies breaking out? Who might be the next “big winner”?

  

• AI's Retention Problem: it’s becoming clear that most LLM-powered application startups are struggling with retention - (1) today’s AI budgets are exploratory, (2) copycats are everywhere, (3) AI apps haven’t offered enduring use cases.

Education and the Future of Work 📚

• To Test the A.I. Learning Hype, I Visited Classrooms: some tech proponents say generative artificial intelligence will revolutionize education. Yet, some schools are blocking it. Across the country, reviews of tutoring bots are mixed.

• EdTech Companies Are Racing to Build a GitHub Copilot for Teachers. This Will Not Be Easy: Generative AI has produced an extremely useful tool for software developers. Can it do the same for teachers?

• As educators embrace AI, debate shifts to bias and equity concerns: some have argued AI, used correctly, can help close equity gaps between students, especially neurodiverse students who may need more specialized plans.

• Teaching History with ChatGPT: using them to simulate interactive historical settings as part of a university assignment.

• AI reading coach startup Ello raises a $15M Series A to bolster child literacy: using its proprietary technology, Ello listens to the child read out loud and analyzes their speech to correct mispronunciations and missed words. Like a real teacher, the AI reading coach waits until the child has finished reading the page before using phonics-based strategies to teach them critical reading skills.

• Delphi announced $2.7M in funding and its new AI digital cloning service: simply upload as few documents containing your communications to it — and as many as thousands, including emails, chat transcripts, even YouTube videos or audio files such as podcasts or voicemails — and Delphi will create an AI chatbot that mimics, as closely as it can, your personality, manner of writing, or speaking, audibly, as of today, through a partnership with voice-cloning startup ElevenLabs.

• Learn.xyz raises $3M Seed Funding: aims to make casual learning accessible and engaging, the app generates quick educational "Curiosities" from any user question, about any topic imaginable.

• MagicSchool Raises $2.4M to Fight Teacher Burnout Using Artificial Intelligence: 100,000 teachers have started using MagicSchool to reclaim their time and better serve students. Over 10,000 schools now have teachers using MagicSchool representing nearly every school district in the country.  

Startups and Tools 🛠️

• Databricks raises a $500M Series I, boosting valuation to $43B: how did Databricks manage an up-round in this market, where more conservative revenue multiples abound? The company said that in Q2, its revenue run rate surpassed the $1.5B mark and more than 10,000 customers globally (of which more than 300 are currently generating revenue at a pace of $1M+/per year for its software and services).

  • Inside Databricks’ Contrarian Playbook: Burn $1.5 Billion to Buy Big Growth

• Imbue raises $200M Series B to build AI systems that can reason and code: their goal is to build practical AI agents that can accomplish larger goals and safely work for us in the real world. To do this, they train foundation models optimized for reasoning and hope to release systems that enable anyone to build robust, custom AI agents that put the productive power of AI at everyone’s fingertips.

• Pixis, an AI-powered full-stack marketing platform, raises a $85M Series C1 round: Pixis develops accessible AI technology for growth marketing and has a rich product suite that includes targeting, in-flight performance optimization and generative AI capabilities.

• LastMile AI closes $10M seed round to ‘operationalize’ AI models: LastMile AI provides the tooling and platform to assist businesses in confidently incorporating AI in their applications - “a single developer platform that encompasses the entire lifecycle of AI app development.

• Lessons from 139 YC AI startups (S23)

  

• GlideAI: a new set of native building blocks that make it easy to create AI-powered apps. No need to manage prompts, choose models, deal with complex API keys, or cache results to optimize cost and performance.

• Epsilon: a faster way to read research.

• Atticus AI: negotiate contracts faster - understand any legal contract, save time and money.

• Boolvideo: turn static assets into cinematic quality short videos.

• Mavex: your personal AI Executive Assistant - scheduling and calendar management, email drafting and beyond.

• TimeOS: your personalized AI for meeting notes & scheduling.

• CustomerIQ: captures every piece of customer feedback around your company, from calls to support tickets, and makes it synthesized and searchable with AI.

Tech 💻 

• MLCommons Introduces New LLM and Recommendation Benchmarks: Nvidia’s H100s performed best on a 6B-parameter LLM that summarizes CNN articles. Intel’s Gaudi2 is second (10% slower than Nvidia).

• Why Nvidia’s AI Supremacy is Only Temporary: (1) Inference will Dominate, not Training, (2) CPUs are Competitive for Inference, (3) Deployment Engineers gain Power, (4) Application Costs Rule.

• LLMs Are Not All You Need: LLMs are great, but when we add prompt engineering, conversational memory, RAG, agents, and guardrails into the mix, they're a lot better.

• LLMs, RAG, & the missing storage layer for AI: beneath the surface of the AI revolution lies a crucial missing element, one that has the potential to unlock even greater AI capabilities: the storage layer.

• Can LLMs learn from a single example?

• Spread Your Wings: Falcon 180B is here: Falcon 180B is the new model in the Falcon series of models from UAE (which are already topping open source charts), and with 180B parameters, it’s now the largest openly available language model.

• Tutorial: Fine-Tune your Own Llama 2

Regulation and Policy ⚖️

• Google pledges $20 million for responsible AI fund: the project will "support researchers, organize convenings and foster debate on public policy solutions to encourage the responsible development of AI."

• The global race to set the rules for AI: companies will need to examine carefully how to operate in specific markets and whether it will require them to design different models or offer different services to comply in a particular region.

• Tech leaders debate perils and possibilities of artificial intelligence at closed-door Senate session: Elon warned of existential threats of AI, whereas Gates highlighted its potential in combatting starvation. Zuckerberg championed the sharing of AI models even with associated risks.

• Pulitzer Prize winning author Michael Chabon and others sue OpenAI: The writers allege that because "when ChatGPT is prompted, it generates not only summaries, but in-depth analyses of the themes present in Plaintiffs' copyrighted works," the writers believe "the underlying GPT model was trained using [the] plaintiffs' works."

  • Artists sign open letter saying generative AI is good, actually

• Microsoft offers legal protection for AI copyright infringement challenges: this new policy, called the Copilot Copyright Commitment, is an expansion of Microsoft's existing intellectual property indemnification coverage.

Other

• A neural network can smell like humans do for the first time: Science Paper, "A principal odor map unifies diverse tasks in olfactory perception".

  

• ChatGPT successfully identifies chronic pain diagnosis 17 doctors missed for 3 years

• (podcast) Sam Altman CEO of OpenAI: ChatGPT, Future of AI, AGI, and China

Resources:

• Celebrating 15 Years of ASU+GSV: Here Comes the Sun! Join us for the 15th anniversary of ASU+GSV to be part of the extraordinary group of educators, entrepreneurs, investors, and innovators reshaping the future through scaled innovations in education and workforce skills at the 2024 ASU+GSV Summit, April 14-17, in San Diego.

• Subscribe to GSV’s N2K Daily Newsletter

• The GSV BIG 10: weekly coverage corner for the top 10 stories, insights, and major plays in learning and skilling.