LLM Links, 2/24

Chris Mims on the future of Web search; Sierra on customer relations; Scott Alexander on a forecasting bot; Ethan Mollick says to skate to where the puck is going

Chris Mims (WSJ) writes,

Over the past week, I’ve been playing with a new, free, AI-powered search engine-slash-web browser on the iPhone, called Arc Search. When I type in a search query, it first identifies the best half-dozen websites with information on that topic, then uses AI to “read” and summarize them.

It’s like having an assistant who can instantly and concisely relate the results of a Google search to you. It’s such a timesaver that I’m betting that once most people try it, they’ll never be able to imagine going back to the old way of browsing the web. 

He speculates that in the aggregate this could backfire. Because you lose the incentive to click through to the site that put of the content, the providers of the content will not be rewarded, and the web will be deprived of content.

Perhaps that will become an issue. But regardless, if more efficient search becomes the main use case for large language models, then I will consider this a major disappointment. “Google on steroids” is not enough to justify the hype and investment in these models.

Yahoo finance has a story on Sierra.ai, a brand new company. Co-founder Bret Taylor says,

If you look at 1995, to exist digitally, you needed a website, 2005 maybe you had a profile page in social media. 2015, maybe you made a mobile app. We believe that your AI agent, an AI version of your company that can simply have a conversation with your customers, may dwarf all of those in terms of their importance to your brand. And, Sierra, we want to be the platform that every company in the world uses to build their AI agent.

The other co-founder, Clay Bavor, says,

what we see it doing today is not just answering questions, but actually taking action on behalf of their customers processing complex exchanges and returns for the retailer, managing subscriptions, and giving advice on things like food points in the case of Weight Watchers. And with Weight Watchers, it's already resolving almost 70% of all customer inquiries and with a remarkable 4.6 out of five star customer satisfaction rating. And so over the next six months, we hope to improve the quality and breadth of the types of actions it can take on behalf of customers. And we're seeing great progress on that front.

Yes. We can’t get replace the “For ___, press 1” customer response systems soon enough. Maybe this company will not achieve its ambitions, but it has the right idea.

Scott Alexander writes,

It knows what Prospera is. It correctly identified the main factor threatening its growth (the ongoing legal case with Honduras). It came up with some really interesting base rates, calculated them, and weighted them for relevance. All of this is great.

He is writing about a forecasting bot created by FutureSearch. What I like is how the bot shows its work.

Ethan Mollick writes,

there is also a lot of value in building ambitious applications that go past what LLMs can do now. You want to build some applications that almost, but not quite, work. I suspect better LLM “brains” are coming soon, in the form of GPT-5 and Gemini 2.0 and many others. When they do, you can swap them into the almost-but-not-quite-working applications for a fast start. This is similar to the philosophy of the big AI labs, which build ambitious solutions (OpenAI's GPT agents, Google's connections to Gmail) which will benefit when the next version of their core LLMs are released.

So don’t just build for what is possible today, but what is possible in six months

Maybe I should start to think about building a virtual version of myself that can go onto a video platform and answer questions from subscribers 24-7.

substacks referenced above:

@

Strategies for an Accelerating Future

@

Mantic Monday 2/19/24