The Verge - Artificial Intelligences 07月18日 20:18
Perplexity’s Comet is the AI browser Google wants
index_new5.html
../../../zaker_core/zaker_tpl_static/wap/tpl_guoji1.html

 

Perplexity公司推出了一款名为Comet的AI驱动浏览器,旨在革新用户上网体验。Comet集成了AI问答引擎,取代了传统的搜索引擎结果展示方式,并提供了一个内置的AI助手,能够与用户进行对话,总结文章、描述图片、甚至处理网页上的特定任务。其核心亮点在于“代理式”功能,即代表用户自动执行一系列操作,如发送邮件、关闭长时间未使用的标签页、发布社交媒体内容、退订邮件以及接受LinkedIn好友请求。虽然部分任务执行效率不如手动操作,但Comet在处理复杂任务如在线购物和预订时展现出强大的自动化能力,并能规避一些现有AI助手的局限性。尽管在某些细节上仍有待完善,如预订信息填写不完整,但Comet已成为挑战Google主导地位的有力竞争者。

🚀 Comet浏览器以AI驱动,将Perplexity的“答案引擎”融入地址栏,取代传统搜索结果,直接呈现相关链接及信息摘要,提供更高效的搜索体验,但可能缺少Google搜索的海量链接选项。

💬 内置AI助手提供类似Gemini的对话交互,可用于总结网页内容、描述图片、处理YouTube视频,甚至扫描所有打开的标签页进行信息汇总和产品比较。

🤖 Comet最突出的功能是“代理式”任务执行,能够代表用户完成如发送邮件、关闭闲置标签、发布社交媒体内容、退订邮件、接受LinkedIn邀请等操作,并能展示其执行过程。

🛒 通过特定指令“take control of my browser”,Comet能更深入地执行复杂任务,如在Amazon上添加商品并完成结账,或处理网页评论区的折叠内容,展现出超越部分竞品的灵活性。

⚠️ 尽管Comet在自动化任务上表现出色,但在处理如餐厅预订等复杂操作时,仍可能出现信息填写不完整等小问题,但开发团队表示将持续优化AI模型以提高成功率。

Perplexity has just launched its agentic answer to Google Chrome — it’s called Comet, and it knocked out a slate of tasks on my behalf, though I think I could’ve done some faster myself. The new AI-powered browser is currently only available to Perplexity Max subscribers or through an early access waitlist, and it’s supposed to simplify the way you browse the web by infusing AI into practically everything you do.

For one, it replaces Google Search results with its Perplexity AI “answer engine,” which appears in your browser window when you type a query into the address bar. Unlike your typical search engine, Perplexity will first surface links to relevant websites and then generate information about what you’re looking for. Comet’s distilled search results come in handy when you want it to narrow down your results for you, but it’s a bit jarring not to see the massive selection of websites suggested by Google. 

Comet also comes with an AI assistant built in, similar to the Gemini integration that Google is testing in Chrome. Selecting the Assistant button in the top-right corner of the browser will open up a sidebar with a chat interface. From here, you can type in a query or use voice mode to chat about different topics, as well as ask specific questions about the webpage you’re on.

Comet can generate a summary of an article, describe an image, summarize YouTube videos, or perform more research about a topic that catches your eye. It’s also able to scan all of your open tabs to provide summaries of those pages and compare products on them.

At this point, these are all pretty standard features for an AI tool, but what makes Comet really stand out is its ability to complete tasks on your behalf. After linking my Google account to the browser, I found that it was frighteningly fast at generating — and sending — an email to myself containing a summary of this year’s hurricane season outlook. The browser also speedily complied with a request to close all the tabs I hadn’t opened in more than 15 minutes. It even wrote and published a post on my X account on my behalf about the upcoming Made by Google event.

I asked it to unsubscribe from the promotional emails sent by Fubo and Fanatics.com as well. I watched as Comet’s AI assistant walked itself through the process. In the chat interface, Comet shows what it’s “seeing” as it locates recent emails sent by the companies, finds the unsubscribe button, and then actually selects it.

I even had Comet go through my list of LinkedIn invites and accept requests from people with five or more mutual connections. The browser once again traced its own process of going through my invites, identifying which ones met my threshold for mutual connections, and then hitting Accept. But as I had Comet perform these tasks, I couldn’t help but think it’d be faster if I did them myself.

It took Comet two minutes to unsubscribe from receiving emails from those two providers, but it only took me a little over 30 seconds to unsubscribe from the same ones (yes, I timed myself). Comet also ate up a chunk of time when accepting a couple of LinkedIn invitations, a task I could do in just a couple of clicks. I can see it serving as a great accessibility tool, as well as a way to complete tasks in the background while you’re doing something else.

You can unlock even more agentic features when you start a prompt with “take control of my browser.” I didn’t realize this until I contacted Perplexity to ask when the browser would be capable of booking reservations or buying products. Without this phrase, Comet will stop short of completing these tasks and instead provide instructions on how you can do it manually. 

To start, I asked Comet to “take control of my browser” and summarize the comments on a Verge article. Instead of denying my request because it couldn’t read the collapsed comments section (like Gemini in Chrome did), Comet worked around this and opened the comments section itself. It summed up the sentiment surrounding my colleague Vee’s cursed piece about Grok’s AI anime waifu, calling users’ reaction to the chatbot overwhelmingly “negative and critical.”

I took things a step further by asking Comet to take control of my browser, add aquarium sand and glue for an iPad repair to my cart on Amazon, and then check out. The process was surprisingly seamless, as I watched it acknowledge the total price, choose Prime’s one-day shipping speed, select my default payment option, and hit “order” without needing me to intervene. 

I only ran into some hiccups when having Comet book me a reservation for a restaurant. When I finally found a restaurant that accepts online reservations, I once again asked the browser to take control and make a reservation for me on a specific date. It completed the task, only it never asked for my email or phone number, and instead entered a generic placeholder for both. I was able to have Comet rebook with my actual email address, but it shows that the browser might not get everything right all the time.

“Some of the more complicated agentic actions like shopping do have a higher failure rate than simpler tasks, but this is actually a limitation of current AI models,” Perplexity spokesperson Jesse Dwyer told The Verge. “So this will only get easier and better in Comet.”

Still, Comet can do far more than Chrome’s Gemini integration, and it’s exactly the type of tool that Google has set its sights on creating. Perplexity CEO Aravind Srinivas has made it clear that the startup wants to challenge Google’s dominance, and Comet may play a big role in bringing it up to speed.

Fish AI Reader

Fish AI Reader

AI辅助创作,多种专业模板,深度分析,高质量内容生成。从观点提取到深度思考,FishAI为您提供全方位的创作支持。新版本引入自定义参数,让您的创作更加个性化和精准。

FishAI

FishAI

鱼阅,AI 时代的下一个智能信息助手,助你摆脱信息焦虑

联系邮箱 441953276@qq.com

相关标签

Perplexity Comet AI浏览器 人工智能 网页浏览
相关文章