What is Solveit? Showing some recent use cases
00:00:00.600 | Hello everyone, Jono here. With the announcement of our new SolveIt course,
00:00:05.360 | people asking what is SolveIt the app, what is it all about, I thought I
00:00:09.240 | should show a few of the cases where I've used it outside of work recently, just
00:00:13.400 | to give you a feel for what SolveIt looks like, the kinds of things that it's
00:00:16.720 | ideally suited for, and what you can do. So SolveIt is kind of like a dialogue
00:00:20.760 | between you, you can put notes to yourself, it's also a place to execute code, right?
00:00:24.800 | It's a Jupyter-style REPL environment, everything is live, so you can see if I've
00:00:29.720 | declared a variable that now exists, and I can modify that, right, and follow
00:00:35.360 | what's going on. And it's also a dialogue between you and the computer, so for
00:00:40.280 | example, you can, any Python tool that you can use, you can give that to SolveIt as
00:00:44.900 | a tool, and it can interact with the code that you've written, variables that you've
00:00:51.260 | defined, and yeah, it's all interacting in that same environment. So you can see
00:00:56.300 | here, it just added to that same variable, it's able to, you know,
00:00:59.440 | tell us how things work, explain things. So some of the use cases, you know, scraping
00:01:06.320 | data from a site, you can see here, like, I'm doing a lot of the work myself, exploring
00:01:12.220 | around, testing things out, but then I'm also able to just hand it like, "Hey, here's
00:01:15.540 | the contents of a page, could you help me pull out the pieces of information that I'm interested
00:01:19.000 | in?" Exploring models, so loading up little models to do bits of ML research on. Again, really
00:01:26.000 | great to be able to just interact, explore, poke at things, and then also if you hit a roadblock,
00:01:30.560 | like at some point something's not working, or you're unsure what's going on, being able to ask the AI,
00:01:34.880 | like, "Hey, you know, how might I check that everything's on the right
00:01:38.780 | device?" Anytime you need that assistant. Yeah, exploring, in this case, some recreational maths,
00:01:44.880 | you know, watching a YouTube video, and then again, like, doing a lot of this stuff myself,
00:01:49.440 | because it's just fun to have a computational playground, as it were, to test that ideas, but
00:01:54.480 | then also being able to ask it to, for example, use the search tool, tell me if I'm on the right track,
00:01:58.480 | it's able to give me, you know, fancy names to put to the concepts that's going on. And also for
00:02:05.200 | things like, you know, in this case, yeah, exploring questions that I have, and writing things like the
00:02:11.120 | plotting code. I don't need to do the plotting myself, I'm more interested in the solving, but
00:02:17.280 | it's really cool to then have something that can say, "Oh, well, here you go, here's your fancy
00:02:21.600 | matplotlib output," right? That's not the end goal for me, but it's nice to have this AI co-worker
00:02:27.760 | buddy. Yeah, so I use this for all sorts of things. I use it when I'm curious whether or not AI can do
00:02:32.240 | something, to pull together some data and send that as an eval. I use it for hopping out, like in this
00:02:37.280 | case, I've got a friend who's a civil engineer, analyzing some data, loading a file, poking at it,
00:02:41.920 | getting info from the AI, how might I pull out the relevant bits, you know, could it write the regular
00:02:46.160 | expression for me? Thank you very much. And then importantly, being able to validate at every
00:02:50.560 | step. Okay, these are the lines I'm interested in. Can I do the convex hull? Yes, it can tell me how to
00:02:54.640 | do the convex hull. Great, I just want the bottom. Yes, it can tell me how to get that. So it's very much
00:02:59.200 | a back and forth between you, between the computer that's executing the code, and between the AI.
00:03:04.080 | So yeah, those are just like some recent dialogues I had open. I hope that elucidates a little bit about
00:03:10.080 | how we use SolveIt as a tool and some of the things it can do. It does feel like a different
00:03:15.600 | way of coding to anyone who's used to the offload everything to the AI. But it also feels like a
00:03:22.720 | very familiar and obvious way of coding if you're at all into the, you know, the way we used to teach
00:03:28.960 | in fast AI, right? The sort of iterative exploratory programming with notebooks and mbdev. And for most
00:03:34.160 | people, it's like, why are you describing this as anything new? This is just how good code is code,
00:03:38.160 | is to do things in small incremental pieces that you understand. But yeah, it's exciting to build a tool
00:03:42.800 | that tries to focus on that in an age where a lot of tools are focused on human replacement.
00:03:46.960 | So I hope you find that interesting. If you want to come check out a course where we'll be covering how
00:03:51.360 | to use this tool and this approach for all sorts of things, check out the new SolveIt course,
00:03:55.680 | how to solve it with code. But if you're just interested in the tool, once that course is complete,
00:03:59.840 | hopefully we'll be opening up just kind of pay-as-you-go subscriptions to access it, to use
00:04:06.160 | our instances in the cloud and our AI. But yeah, otherwise you can take these ideas and you can
00:04:12.000 | apply them anywhere. So yeah, I just hope you get some encouragement from this to keep on playing
00:04:17.280 | playing with code and working at your craft.