??ChatGPT search, ??Claude Computer Use, ??Understanding agents better
OpenAI released ChatGPT search late last week for Plus and Team Users. Now you can search directly from the ChatGPT and get up-to-date info with easy links to sources.
The early impressions are solid. It's useful, fast, and mostly accurate. There are no ads in ChatGPT search results, yet. It’s better than Google and Bing for search result quality in some areas (time sensitive queries and recent sources) and isn’t as good at local search.
ChatGPT isn’t much different than Perplexity and Bing on the surface, but it puts AI-powered search in front of OpenAI’s 200 million weekly active users.
Gartner predicts search engine volume will drop by 2026 due to chatbots and agents. If you're not using AI-based search already (ChatGPT search, Perplexity, etc.), it's time to learn.
Claude Computer Use analyzes construction site
A week before ChatGPT search, Anthropic gave Claude the power to use computers.?
You can see Ethan Mollick testing it with a complex video analysis task below. Professor Mollick gave Claude a YouTube video of a construction site and prompted it to analyze the site. Claude 3.5 Sonnet with computer use, watched the video and took screenshots every few seconds to “study” them, so it could analyze the construction site.
After studying the video Claude created a spreadsheet summarizing what it observed in a few seconds. It outlined all the issues and made recommendations for fixing them -- including identifying exposed rebar without caps.
"The results seem good from reviewing the video, but I am not an expert, and I would be surprised if there were not serious hallucinations mixed in. For this and many other reasons, I would never want this system to be used to punish or reward people." Ethan Mollick -Associate Professor at The Wharton School
Understanding agents better
While Claude is using computers, Microsoft Copilot gave developers the choice to use models other than OpenAI LLMs. Now you can use Anthropic's latest Claude 3.5 Sonnet alongside Google's Gemini 1.5 and compare them in coding tasks to OpenAI o1-preview.
While we have more choices for the LLMs behind agents, technologists are starting to discover and map the ways you can get your agents to work together.
Here's an example of an agentic patter you might use for a customer service chatbot:
"The Collaboration Pattern with Three or more Agents"
Interesting? The diagram above comes from an overview of different agentic patterns. Definitely worth a look if you're building anything based on agents or want to understand how to.
There are a lot of interesting links this week??
Useful Links & News
领英推荐
AI Research Papers & Technical Links
PS
We hope these newsletters help you stay informed. We want to know what you think is useful or hear about news you're excited about. Tell us in comments below!