🚨 OpenAI Releases Search and Claude Releases MacOS app
October 31, 2024
Two big AI drops landed recently: OpenAI’s new search feature and Claude’s Mac app. Here are my first impressions, what to watch for, and how this might shape your workflow.
OpenAI Search: first impressions and takeaways#
- What I saw
- OpenAI surfaced a dedicated search-like tool (coach tip) inside the interface. Clicking it leads to a set of search options that feel similar to other multi-source AI search experiences.
- The experience emphasizes source transparency: results show multiple sources with badges, which helps you assess accuracy and bias.
- How it works under the hood
- This is an agent-based search approach: multiple models/prompts are used to fetch and rank results, then surfaced with sources. It’s reminiscent of Morphic’s “system of chaining agents.”
- In practice, you type a query and the system returns a mix of sources and a summarization style, with the ability to compare against other results.
- A couple of concrete experiments
- Trump garbage truck (news/event framing): Top results pull in multiple outlets (AP, NY Post, etc.). You can see sources, which supports trust and helps you verify context.
- Weather forecast (non-political): The tool returns multiple results with source badges; you can choose which result to trust and even compare with other sources.
- Overall pattern: The tool pushes for multi-source results and a more nuanced answer rather than a single paragraph. This pressures single-model, one-shot summaries and nudges toward richer, sourced outputs.
- Observations and takeaways
- This approach puts significant pressure on standalone chat-centric search players (like some configurations of Perplexity) to innovate beyond short summaries.
- Source transparency is a real win for trust, though some queries (e.g., hurricane specifics) may still require cross-checking with dedicated sources.
- The experience is promising, but performance and factual depth will determine long-term value. Expect ongoing refinements as OpenAI polishes the agent-chaining and source integration.
Claude Mac app: first look#
- Availability and setup
- Claude now has a Mac OS app (beta) for a dedicated desktop experience, rather than relying on a saved Safari/web workflow.
- It’s designed to be fast to launch with hotkeys, and you can customize the key bindings in Settings (Command +, to open, etc.).
- What the app delivers
- Clean, focused UI with standard project management hooks (Projects, etc.). It’s designed to be a self-contained Claude workspace.
- You can dictate messages (iPhone voice input integration is mentioned) and you get the usual Claude capabilities in a desktop context.
- Copilot integration is highlighted as part of the broader ecosystem story.
- Usability notes
- The app is described as “beta” and, like many early desktop AI clients, can lag behind a best-in-class web experience in some scenarios.
- Performance and memory usage are typical concerns for Electron-based apps; the presenter noted it’s an Electron app and demonstrates the usual trade-offs.
In practice, the app is a strong productivity play for Mac users who want a dedicated, keyboard-friendly interface with quick access via hotkeys.
- Limitations and caveats
- Web-based Claude interactions may still feel smoother for some tasks; the desktop app’s advantage is dedicated focus and hotkeys, but web might outperform it in certain prompts or workflows.
- Some interface experiments (like deeper dev tools inspection) are restricted in the app, which is common for Electron apps but worth noting for power users.
Practical takeaways for your workflow#
- For end users
- If you’re on Mac and you rely on Claude, try the Mac app for a focused, hotkey-driven workflow. Monitor RAM/CPU usage and compare responsiveness with the web version for your typical prompts.
- When evaluating OpenAI’s search feature, pay attention to sources and provenance. Use queries that require cross-source validation to gauge usefulness and trust.
- For builders and buyers
- Agent chaining and multi-model orchestration are clearly gaining traction. If you’re building search or QA tooling, consider how you expose sources and allow users to pick or compare sources easily.
- The app vs. web trade-off matters: desktop apps can boost focus and speed but may lag behind the web in terms of feature parity or dev-tools access. Design around those trade-offs.
- For content creators and learning
- Expect rapid shifts in how these tools surface sources and how they handle long, structured outputs (events, summaries, reactions). When you cover these topics, highlight source provenance and model behavior, not just the surface answer.
What’s next#
- Parker’s approach: continuing to publish high-quality, long-form explorations of AI tooling. We’re moving toward deeper, weekly episodes that dissect what’s working, what isn’t, and how to integrate these tools into building real products.
Links#
- OpenAI ChatGPT Search — official announcements and docs
- Claude Mac app — Anthropic/Claude blog updates and Mac app release notes
- Perplexity — reference for comparison of search and summarization UX
- Morphic — project focused on agent-based reasoning and chaining agents
If you want deeper dives, I’ll be unpacking these tools further in upcoming episodes, including hands-on tests with real prompts and integration workflows for my SAS work.
Transcript
pretty good Halloween for the folks in the AI scene and we got two new drops that came out you'd only know about one of them if you were not on a wait list with open a ey but in this video I'm going to be walking you through my first impressions of the new CLA app what it comes with this is available on Mac OS as of earlier today from a post from the CPO and previous co-founder of Instagram where he mentioned that it's out and then the search feature for open AI so what I've done here is this was the first suggestion on open AI what happened was I opened up open a ey and I get this nice little tool tip the coach and if I were to click on that then it'll show me a couple of these options you'll notice this is really reminiscent of perplexity and what I'm curious about here is how will perplexity hold up over time what is their secret sauce if if you've watched my previous videos you know that I've talked a lot about how there's a great project called morphic and it basically rebuilt perplexity and broke it down and a lot of the magic is in the system of chaining together agents so the way you can think about it is type something in in here and that's just a oneshot answer whereas perplexity you string together a few of the different agents and without getting too nerdy on it let me just see if I can can't find a little something to show you so if I type in writer in here I should be able to find something in here so you can see there's an issue or a document around the writer agent model so he's chaining together and how this plays into perplexity is they're chaining together different agents that have different models that they can use and different prompts and they're chained together so when I type something in like on GPT if I click Trump garbage truck it went through it selected the sources now what I find weird is I couldn't really see for that long where the sources came from and I think this is really important they're trying to take up the most valuable resource on the internet the most valuable you know company is or one of them is is Google and that's search it's sorting your information and Google has kind of dropped the ball and I hate to say that on here because I'm on one of their platforms I don't want them to hate on me but if I type in Trump garbage truck it looks like it's matching pretty closely if we look at the top results it's you know Trump hits Wisconsin it's this photo this photo could be from a different Source but it's Associated Press News New York Post and the Australian regardless of your politics it's good to know the sources because there's different leanings especially for something I find it really interesting that they made this the one and I think it's perfect um but you can see all these sources and if I compare that where if I type in Trump garbage truck then I can see it's pretty similar and I recently made a post as soon as I got this email because I'm waiting for someone on a call where you kind of compare the two you can see they're very similar it looks like perplexity is more focused on kind of giving this one paragraph summary and all that's programmatically said obviously and then the key details of the event so it knows it's an event has the context of that and then it kind of breaks it down has reactions so looking at them side by side it's going to be interesting to see how perplexity continues to innovate I think if you just looked at this example you would probably say that perplexity is doing a better job but it's tough because I don't think they have foundational model they don't have the the secret sauce underneath it that's actually powering all this so let's try another one I'm going to use web search and I'm going to try something a little less political so let's just ask what is the weather forecast in West Palm Beach over the next 14 days and is there a hurricane coming so let's take that query and oh cool it looks like it's split it and I've got two now I've got two and I can see that this is cut off the bottom so let me adjust that can see I got two and I can select which one I want there's these tags these badges where I can see the source of that information oh sorry I'm still dictating if you're curious what that is it's just the built-in ma dictator and then I can say well you know weather should be pretty objective it's a number you can stick to it there's no variety here except one is without a nested breadcrumb and one is with but really the same thing so I'll just say I prefer that now let's drag this up and it says I prefer that now let's see if we just move perplexity down here and we paste in a new query we say that oh cool so they they actually give this a little something more um more Dynamic pretty cool looking doesn't say anything about a hurricane interesting so let's dig in deeper we can check so typically I know that this is not where you would go to find a hurricane so for a natural disaster you'd probably want to know that there's these two things out here so let's compare that against Google I mean that's just night and day right so it's becoming pretty clear that the agentic search is going to be very helpful um now what it didn't do is it it didn't tell me anything about the actual hurricane which is not great um and if I went in and dig dig around here you'd know that there's something Brewing so that's the first one that's search and my take away from it is it's going to apply a significant amount of pressure to perplexity and consumers ultimately benefit from this so if you look at the perplexity documentation if you're a developer you can skip ahead if you're not a developer but it just kind of shows you this is is pretty Barren you have chat completions um you know you can you can have it go out and return all that stuff that you just saw and they suggest that you use llama but you know there's not a ton to it so that's the first one the second one we can go ahead and look at the anthropic Claud blog and let's open up nope that's not it artifacts are now generally available we should be able to get to the news um I don't see anything in here which is kind of weird but yeah they came out with a Mac app and why that's helpful is if you're a Mac User you know that it's kind of annoying trying to use CLA with a saved Safari app and what I also think is interesting is we're kind of in this this battle between the bars the the floating command pallets where you have Claud which this is what it looks like so I have a h set to key binding of shift command left bracket for sorry for perplexity and then shift Command right bracket for CLA and I can just start typing in there they do denote that it is in beta and it just does what you'd think so it's not going to do very well with web but this is what the application looks like and by default it has a different binding but you can change that by going into settings if you click command comma and then you can clear that so we know that web stuff isn't great for this what's nice about it is it seems pretty pretty good um what I've noticed with the chat GPT app is it actually is much worse than the web experience so again this has only been out for a few hours but having really the only benefit I do see is that it's dedicated and that you have the hotkey so when I hit that hotkey and it's already focused it just refreshed the page but I think it'll be nice to have this as its own application I hope to see a little little bit more out of it um I hope that it doesn't lag a ton but it does have all the normal stuff that you would expect you have your projects and all that um let's see if there's anything that we can dig into about their announcements so it looks like you can dictate messages to Claude on the iPhone now great I thought that I was already out um it looks like cla's now available on the co-pilot that's great for co-pilot seeing a lot of stuff about that going have to test that soon and people are asking you know what are the benefits of using this significantly easier to access Okay computer use that doesn't support it yet right voice mode okay won't take more RAM it is an electron app interesting I you know I didn't catch that so what that means is I should be able to inspect the elements typically on an electron app if like for instance if I open notion calendar which I know is an electron app then and this is like a dummy calendar this is what I do every day I grind but I can do uh command shift I and then I can actually see what's going on underneath the hood so I can see all the network calls so like if I refresh this is what you would expect to see from a normal website because it's an electron app but it seems like they disabled that so if I do command shift I I can't inspect what's going on which is probably on purpose they don't want you to be able to Snoop around and see how they're sending Network calls maybe there's things in there they want people not to see but that wouldn't make sense because I can just do that in a browser so that's all for now if you are interested in this AI stuff this is an emergency video I wanted to get this out since it came out just a few hours ago but every single week where I break down all the things that I'm learning as I build my SAS and I'm doing it in public and I walk through pretty much you know I'll probably talk about this again tomorrow in more detail but there are dozens of these findings that I come across each week and the issue is for you the viewer there's so much um just bad content out there so instead of making a lot of small videos about here's this little tiny thing I will be from now on making higher quality longer form videos where you know when you see the Parker X show and a new episode comes out every single Friday that you need to tune in if you want to know what's going on with AI what's going on with development tooling because it moves very quickly I would dare to say that I know more about this than most because I spend upwards of 16 hours a day working on this whether it is for my SAS product or doing consulting work to fund the SAS product and I learn a ton of stuff and as I go I'm filtering out the bad stuff thanks again for watching go crush your goals