Hello everyone, welcome to TechCrunch's regular AI newsletter. If you'd like to have this sent to your inbox every Wednesday, sign up here.
Thanksgiving is almost here, and you know what that means. This means stuffing your face until you're in a coma. Well, that and having to entertain all those obnoxious relatives you've managed to keep at bay for the rest of the year.
For those of us lucky (unlucky?) who have been given the task of deciding on this year's feast, there isn't much time left to decide on the menu. Deciding what to prepare is always a difficult decision. Here are some smart ideas. Ask the chatbot a question.
Yes, yes, this has been done before. Turning to AI for desperate Thanksgiving help. (The New York Times tried ChatGPT recipes in 2022.) The results were middling overall. But perhaps the prompt was the problem.
Curiosity got the better of me. So I asked ChatGPT and Claude, two of the more popular chatbots, to come up with a Thanksgiving menu that would be “surprisingly unique for any family member.” I thought that was enough.
Dear reader, AI did not disappoint.
ChatGPT recommends starting with cocktail hour — sumptuous! — featuring whipped sweet potato and goat cheese crostini. Meanwhile, Claude shot for the moon and suggested an appetizer called “Butternut Squash Bisque with Sage Bubbles,” which certainly ticked the “unique” box.
Does “Pumpkin Soup Shooter with Cinnamon Crème Fraiche” sound good? This is what ChatGPT suggested for an appetizer, followed by a main course of miso butter turkey with ginger soy sauce glaze. Claude, once again a wild card, suggested “Lavender and Fennel Cured Turkey with Honey Thyme Glaze.'' The chatbot described it as a herbal departure from the classic roast turkey. surely.
What about the sides? ChatGPT recommended the chili lime cornbread and pistachio risotto. Claude said he would whip up this premium liquor to make “wild mushroom and chestnut stuffing and aged sherry.”
To finish off, both chatbots ask you to stick to classic menu items like pie, cheesecake, and healthy ice cream. The twist? The ice cream is flavored with saffron, and the cheesecake is flavored with chai spice.
“This menu takes the familiar flavors of Thanksgiving and elevates them with unexpected ingredients, techniques, and combinations,” Claude writes of its creation. “Each dish tells a story and sparks conversation, allowing us to share not just the food, but the experience and creativity.”
There's no arguing with that. But as this year's designated chef… well, I'll just say that I'm not aiming to be a top chef.
news
OpenAI's Sora leaked: A group appears to have leaked access to OpenAI's video generator Sora in protest of so-called “artwashing” on OpenAI's part.
Amazon backs Anthropic again: Anthropic has agreed to raise an additional $4 billion from Amazon and train its flagship generative AI models primarily on Amazon Web Services, Amazon's cloud computing division.
AI app connectors: In other news from Anthropic, the company proposed the Model Context Protocol, a new standard for connecting AI assistants to systems where data resides.
OpenAI funds 'AI morality' research: OpenAI is pumping $1 million into a Duke University research program to develop algorithms that can predict human moral judgment.
YouTube gets AI backgrounds: YouTube's Dream Screen feature, YouTube's short-form video format, now allows users to create AI-generated video backgrounds.
Brave adds AI chat: Search engine Brave has introduced an AI chat mode for follow-up questions based on the initial query in Brave Search. This is an extension of Brave's AI-powered Answer, which provides an AI-generated overview of web searches.
Ai2 Open Source Tülu 3: The Allen Institute for AI (Ai2) has released Tülu 3, a generative AI model that can be fine-tuned and customized for a variety of applications, such as solving math problems.
Crusoe raises funding: Crusoe Energy, a startup building data centers reportedly leased to Oracle, Microsoft and OpenAI, is raising $818 million, according to an SEC filing.
Threads tests AI summaries: Meta's Threads has started testing AI-generated summaries of what people are discussing on the platform, inspired by Rival X's page.
This week's research paper
Image credit: Google DeepMind
DeepMind, Google's AI research organization, has developed a new AI system called AlphaQubit that it claims can accurately identify errors in quantum computers.
Quantum computers have the potential to be much more powerful than traditional machines for certain workloads. However, it is also prone to “noise”, or common errors.
AlphaQubit helps identify, mitigate, and correct these errors, making quantum computers more reliable.
However, it's not a perfect system. Google acknowledged in the post that AlphaQubit is too slow to correct errors in real time and is not particularly data efficient. The company says it is working on an improved version.
this week's model
Sample Runway Frames model. Image credit: Runway
Runway, a startup that builds AI tools for content creators, has released a new image generation model that it claims offers greater style control than most others.
The model, called “Frames,” is being gradually rolled out to users of Runway's third-generation alpha video generator, and Runway says it can reliably create images that are true to a specific aesthetic.
Now, it's worth noting that Runway may be playing fast and loose with copyright rules. A report from 404 Media earlier this year suggested that the company scraped YouTube footage from channels belonging to creators like Disney and MKBHD without permission to train its models.
When asked for comment, a Runway spokesperson declined to reveal the source of Flames' training data.
Like many generative AI companies, Runway claims that its data scraping practices are protected under the fair use doctrine. This theory has been tested in numerous legal battles, including a class action lawsuit filed against Runway and several of its Art Generator rivals.
grab bag
Nvidia has announced what it calls “the world's most flexible sound machine.”
The chip giant's model, called Fugatto, can create music, voice, and sound combinations from text descriptions and a collection of audio files. For example, Fugatto lets you create musical snippets based on prompts, remove or add instruments to songs, and change the accents and emotions of your vocal performance.
Nvidia says Fugatto has been trained with millions of openly licensed sounds and songs, and can generate things that don't even exist in the real world.
“For example, Fugatto can make trumpet calls or meow on a saxophone,” the company said in a blog post. “Researchers have discovered that with some fine-tuning and a small amount of singing data, it can handle tasks that it cannot handle. [trained] For example, generate high-quality singing voices from text prompts. ”
Nvidia has not released Fugatto due to concerns that it could be exploited. However, Reuters reports that the company is considering how to launch the model “responsibly”.