AI
AI Searches Just Got Way Faster
If youâve been using Perplexity or any other AI-based search engine, you know that it takes a few seconds to get the answer to your questions typically. Thatâs why I was so psyched about Perplexityâs announcement today that they had shipped an updated version of their in-house Sonar model built on Llama and powered by new Cerebras infrastructure. One of the reasons that Perplexity is so great is that they use specialized AI models tuned specifically for search. Sonar is much faster, while still providing great search results. I highly recommend you check out their blog post about the changes that theyâve made and give the updated search tool a try. I was pretty blown away, particularly compared to ChatGPT Search. In my tests, Perplexity with the new Sonar model was three to five times faster than ChatGPT Search. Typically a search through Perplexity would surface links instantly and output generated in about two seconds. When I would run the same searches through ChatGPT Search, it would take typically six to ten seconds to get the entire result. I also vastly prefer the output from Perplexity, the tuning they’ve done seems to make a big difference both in the substance and the sources used. What they’ve accomplished is really astonishing.
Operator May be The Future, But itâs Still Too Early
A little over a week ago I reluctantly paid $200 last week to try out OpenAIâs new Operator web app designed to allow you to automate complex tasks. The tool, a new website included in the ChatGPT Pro subscription offering, lets you command a virtual machine to go out and complete requests via a browser. Itâs similar to what Rabbit created for their LAM feature for the R1 gadget, if youâre familiar with that.
Upon opening Operator, youâll see a range of suggestions like trip planning, food delivery, and restaurant booking. Those are all compelling things to automate, but theyâre not necessarily things that average people do constantly. The fact that OpenAI sees these as Operatorâs strengths at the moment might help in-part explain why itâs gated to the $200/month subscription. While I canât exactly get myself to aggressively integrate Operator into my workflow, particularly because I am not comfortable signing into accounts on remote virtual machines, I can see how it is the future. I would love to send my Operator out to clean up my inbox, to beautify presentations, to organize paywalled stories across sites, to post on all social media sites for me at once, to clean up storage space on my machines, to create watchlists on streaming platforms, and so on. The possibilities are endless, but the reality is that we need Operator to run locally or a formal API that securely interfaces with these services. I do not want a remote browser controlled by someone else to have carte blanche access to my most important accounts. Without granting Operator those credentials, it isnât particularly useful.
Thereâs another piece of the puzzle worth touching on, which is that Operator is just not quite necessary yet for me personally. I quickly realized that I just donât need it, despite how amazing it is to play with. Even some of the tasks I described above as use cases that make sense for me personally are things I often enjoy doing on my own. It also happens to just not be quite ready. While playing with it, Operator would occasionally get lost or confused when browsing certain sites. OpenAI has acknowledged this is the case, but I still fear it doing something it shouldnât. Weâll see how it evolves over the next few months. Ultimately, once Operator is part of a standard ChatGPT Plus subscription, it may be more palatable. But for now, you really donât need to spend the money to try it.
Perplexity Pro Now Has R1, o3-mini, and Just About Everything Else
Iâve used nearly all of the premium AI subscription services, but none of them come close to offering the value that Perplexity Pro does. Despite the companyâs contentious relationship with publishers, Iâve continued to believe it is one of the best-positioned startups in the space. After the emergence of DeepSeek, my lack of usage of Sora or Operator, and realizing my primary use case is almost always search, I moved on from services offered directly by the LLM providers.
Perplexity is particularly appealing because of how quickly the company moves to integrate new models and features. I was floored by how quickly they had DeepSeek R1 up and running both on-site and in their apps. No one capitalized better on the DeepSeek moment in the US than they did. Fortunately, this wasnât a one-off. When OpenAI surprised us all with o3-mini on Friday, I was anxious to play with it. Of course, I could do so because my ChatGPT subscription was still active. But it took a day for Perplexity to have o3-mini integrated into their reasoning system right alongside DeepSeek R1. That means that Perplexity now has every major model (sans Llama, though that might be a benefit given how itâs inescapable inside of Meta apps already) built-in to their Pro subscription. Users get 500 reasoning queries a day using R1 and o3-mini, safely hosted in the US, coupled with the best web search offering there is.
That would be enough, but recall that Perplexity Pro already had several advanced models built-in. Your subscription gets you access to Claude 3.5 Sonnet, GPT 4o, and Grok 2. Plus, you get access to Perplexityâs own models. Instead of having to subscribe to Claude or ChatGPT, you can get them both in a single subscription. And it precludes you from having to use the official Grok or DeepSeek solutions. Again, all of this would be enough if you didnât also get Playground v3, DALL_E 3, and FLUX.1 for image generation. Itâs a mind-blowing amount of value for a single $20/month subscription, which would otherwise cost you the same to subscribe to just one of some of these.
Perplexity ties it all up and adds a nice little bow on top. The UI of the app and the website are great and continue to get better. They keep adding data widgets, switching to a writing or math focus is easy as can be, and you can even build out dedicated pages for topics and edit them as you wish. Itâs a ridiculously good offering and if you have an Android device, the assistant just makes it all that much better. I canât recommend it enough. If you havenât used Perplexity, try the free version and I think youâll immediately be blown away by how good it is as a search engine. But if you want to get more out of these tools, itâs well worth trying the Pro subscription.
(Perplexity now has Gemini too as of 2/5!)
Perplexity Assistant is the Best Example of Apple Needing to Expand Extensibility on iOS
Last week Perplexity launched their latest product: Perplexity Assistant for Android. I’ve been a big fan of Perplexity for a long time. Despite its often contentious relationship with publishers, they make a spectacularly good tool. I love them, if only for the fact that they’re the first search engine to give Google a run for its money in years. But this isn’t about the search engine game or the AI search race. Perplexity Assistant is a replacement for Google Assistant and Gemini on Android phones. It can be summoned from anywhere, understand screen context, answer complex questions, and do normal personal assistant things like set reminders. I’ve been using it on my Pixel 9 and have been extremely impressed. It’s created an even starker contrast to the experience I have on my iPhone. Just the other day, John Gruber wrote about how the latest incarnation of Siri can often be worse than the previous non-AI infused version. Despite Siri’s new AI features, I still mostly use it for reminders, alarms, and calendar events. I almost never use it for complex questions or even for the ChatGPT integration. I prefer to just use the native app. Perplexity Assistant on Android is so good, both functionally and visually. It’s often better than Gemini and it is leaps and bounds ahead of Siri. It is the first example in my opinion, of how much we could benefit from Apple opening up more parts of iOS. I continue to believe that Apple should not wholesale open up the operating system to work like Android. But in this new day and age of artificial intelligence, it would be a great reward to users to allow us to use other assistants. And I know what you’re thinking, wouldn’t this just highlight how bad Siri can be? Sure, but I think it would also further light a fire under Apple’s butt to fix Siri. Real competition to drive improvements. They can’t coast.
DeepSeek Arrives at an Awkward Time, But itâs Still Amazing
If you havenât tried DeepSeek yet, go do it right now. The app, which is a Chinese-owned AI tool designed to compete with the likes of ChatGPT, Gemini, Llama and Grok, has rocketed to the top of the App Store charts recently and for good reason. DeepSeek is special, not for its user interface or general experience, but because it shows you how it âthinks.â The company trained its model differently; frankly, I donât fully understand how, but I trust that they did so in a way that makes it better than many other open-source models. While DeepSeek is rising to prominence at an awkward time, look no further than the TikTok divestment situation; it is well worth a try. At this point, I am still using ChatGPT as my primary AI tool. But being able to watch DeepSeek think through a request and show you exactly how it got its output is extraordinary. Oh, and the best part? Itâs completely free, yet it rivals the best premium models from other companies. If you donât want to use DeepSeek directly for privacy or censorship-related reasons, you can also do so locally on your Mac using Ollama or on iOS using the latest TestFlight beta of Fullmoon from Mainframe.