AI Week: OpenAI Attracts Anthropic and Musk’s Rich Rivals

Image credit: aap Arriens/NurPhoto/Getty Images

Keeping up with an industry that is changing as rapidly as AI is a tall order. In the meantime, here’s a handy wrap-up of last week’s articles in the world of machine learning, along with notable research and experiments we didn’t cover alone.

The biggest news of the week (I politely left the Anthropic story out of consideration) was Bedrock’s announcement. This is an Amazon service that provides a way to build generative AI apps via pre-trained models from startups such as AI21 Labs, Anthropic, and Stability AI. Bedrock, currently available in “limited preview,” also provides access to Titan FM (Fundamental Models), a family of AI models trained in-house by Amazon.

It makes perfect sense that Amazon would want to have a horse in the generative AI race. could be worth more than $100 billion.

But Amazon has motivations beyond grabbing a slice of the growing new market.

In a recent Motley Fool article, Timothy Green provided compelling evidence that Amazon’s cloud business may be slowing down. The company reported a 27% year-over-year increase in cloud services revenue in the third quarter of 2022, but the rise slowed to the mid-20%. Rates until the end of the quarter. Meanwhile, Amazon’s cloud operating margin declined 4 percentage points year-over-year in the same quarter, suggesting Amazon expanded too quickly.

Amazon clearly has high hopes for Bedrock, even going so far as to train the aforementioned in-house model before launch. And lest anyone question the company’s seriousness about generative AI, Amazon isn’t putting all his eggs in his one basket. This week, CodeWhisperer, a system for generating code from text prompts, became free for individual developers.

So, will Amazon acquire a meaningful slice of the generative AI space and reinvigorate its cloud business in the process? Ultimately, there will come a time when the chaos of generative AI will subside and competitors big and small will emerge.

Other notable AI headlines from the past few days include:

• The wide, wide world of AI regulation: Everyone seems to have their own ideas on how to regulate AI. So there are around 20 different frameworks in every major country and economy. Natasha delves into this full (as of now) list of regulatory frameworks (including outright bans like her ChatGPT in Italy) and their potential impact on the AI ​​industry where they exist. But China is doing its own thing.
• Musk challenges OpenAI: Not content with dismantling Twitter, Elon Musk is reportedly planning to take over his former ally OpenAI, and is now trying to raise the necessary funding and personnel to do so. . Busy billionaires may tap into the resources of multiple companies to accelerate their work, but there’s good reason to be skeptical of this effort, Devin. write.
• Elephants in the room: AI research startup Anthropic aims to raise up to $5 billion over the next two years to compete with rival OpenAI and enter more than a dozen major industries, according to company documents obtained by TechCrunch. In a document, Anthropic says it plans to build a “frontier model” (provisionally called “Claude-Next”) that is ten times more powerful than the most powerful AIs today. This will require $1 billion in spending over the next 18 months. .
• Build your own chatbot: An app called Poe will allow users to create their own chatbots based on prompts combined with existing bots such as OpenAI’s ChatGPT. First publicly available in February, Poe is the latest addition to Q&A site Quora. Quora has long provided answers to the most Googled questions for web searchers.
• Beyond Diffusion: The diffusion models used by popular tools like Midjourney and Stable Diffusion may seem like the best we have, but the next one is always coming. He’s an order of magnitude faster than the likes of DALL-E, Devin reports.
• Small town with AI: What happens when you fill a virtual town with AI and unleash it? Researchers at Stanford University and Google set out to find out in a recent experiment on ChatGPT. Their attempt to create a “true simulation of human behavior” was undeniably successful. The 25 AI interactions powered by ChatGPT were compelling and surprisingly human-like.

• Generative AI in the enterprise: With one piece of TC+, Ron writes about what an innovative technology like ChatGPT can do when applied to the enterprise applications people use every day. Getting there, however, will require creativity to design new AI-powered interfaces in sophisticated ways so that they don’t feel bolted on, he notes. To do.

---

Other machine learning

Meta has open-sourced a popular experiment that can animate even the crudest portraits of people. This is one of the unexpected applications of this technology that is both fun and downright trivial. Still, people liked it so much that Meta made the code free to run so anyone could incorporate it into anything.

Another meta-experiment called Segment Anything had a surprisingly large response. LLM is so hot right now that it’s easy to forget about computer vision. Even then, it’s about certain parts of the system that most people don’t think about. But segmentation (identifying and outlining objects) is a very important part of any robotics application, and as AI continues to permeate the ‘real world’, it’s becoming more important than ever that AI can segment anything. has become important.

Professor Stuart Russell has graced the TechCrunch stage before, but our 30-minute conversation only scratches the surface of the field. Luckily, he regularly gives lectures, talks and classes on this topic. Even with a provocative name such as “How not to let AI destroy the world”, he has been familiar with the topic for a long time, which makes it very grounded and interesting.

Check out this recent presentation presented by another TC friend, Ken Goldberg.