Sunday Links: $6B for another foundation model, bots, and more publisher licensing

It's a little late this weekend, but maybe this will help you kick off your Monday. Here are this week's links:

• Perplexity is rolling out report creation. Perplexity is still my go-to web search tool despite Google bringing Gemini into play for search summaries (see below on how that is going!). The company is now releasing a special mode to create a research report on any topic guided by the audience's expertise (anyone, beginner, expert, etc.). This is a logical product extension that takes perplexity into a space that is more about "research" than it is about "search." The big leap would still be making these searches/research spaces collaborative, but this is a nice step forward.
• ElevenLabs debuts AI-powered tool to generate sound effects. Elevenlabs has thus far focused on human voice models, but this week, it released a sound effect generator. The tool looks pretty cool (though there are already a few of these in the world). What I found interesting is that the press release seems to indicate that the model was trained on Shutterstocks's sample library. This seems like a challenging commercial deal to do - surely creating unique samples from scratch is a major risk to Shutterstock's core business. Why would one go to Shutterstock to browse the list of sizzling steak clips when it's possible to create a unique one using AI?
• Bots are the next level. Anthropic announced this week the ability of its LLMs to make callouts to specific tools pre-specified by the users. These tools are API calls to a back-end service, such as purchasing an item, accessing email, or something else. This isn't surprising, and it is available with ChatGPT in theory with the marketplace of GPTs. However, it's a reinforcement that the endgame of AI/LLMs isn't to answer questions and summarize data; it's to take actions on behalf of the user. Another important step will be having the bot running 24//7 and taking actions based on triggers, such as the ticket price for a concert dropping below a certain level.
• More publisher licensing deals for OpenAI. The Atlantic and Vox Media both make licensing deals with OpenAI. In exchange for accessing their archive for training and real-time content returns (with citations), the companies can use OpenAI's AI to infuse it into their content creation and management process (and perhaps there is a financial component that has not been disclosed). It's interesting to see that OpenAI has the ability to "pay for content" by offering the use of its tools "in kind" as part of payment. That could turn out to be a significant advantage in the fight for training content.
• Elon Musk’s xAI raises $6 billion to fund its race against ChatGPT and all the rest. It would have been interesting to be a fly on the wall for these valuation discussions ($24B post-money). The general consensus over the last few months on LLMs seems to have shifted to the point that investors expect only 5-6 large models to survive, whereas others target niches. Meta's continuing releases of better and better Llama models mean that it will be hard for LLM builders to really differentiate and monetize. How will X.ai do that? I guess the consensus answer is to "always bet on Elon" and the approach does seem to be to be less restrictive than other large models. Rebells get to make more mistakes than incumbents.

Wishing you a wonderful week!