WEBVTT 00:00.160 --> 00:03.280 And so I hardly need to tell you your challenge for this week. 00:03.280 --> 00:09.920 Your challenge is to make the sidekick your own, build it out, add in tools, add in some tools that 00:09.920 --> 00:16.720 are actually specific to things that you do that are able to to give it powers to do things that you 00:16.720 --> 00:17.400 want to do. 00:17.840 --> 00:21.320 Now, some other more fundamental ways that you can improve it. 00:21.360 --> 00:28.240 One of them that I realized is a note to take from the way that OpenAI does theirs is we could encourage 00:28.360 --> 00:33.240 the, the, the llms to start by asking clarifying questions. 00:33.240 --> 00:38.840 You could have it so that the first thing it does is it asks for three clarifying questions, and then 00:38.840 --> 00:41.720 it uses them as part of its ongoing work. 00:42.160 --> 00:44.120 So that would be a really important change. 00:44.440 --> 00:47.960 Another change would be to build out the graph a bit more. 00:48.080 --> 00:53.520 We've put a lot of power into that one assistant, and the problem with that is that we're expecting 00:53.520 --> 00:54.280 a lot. 00:54.320 --> 01:00.120 We're expecting one model to to be able to to take a lot of different tools and be very flexible. 01:00.320 --> 01:03.760 And the danger is that it loses some of its abilities. 01:03.800 --> 01:09.720 It loses some coherence because the context is so long, because there's so much it can do that it gets 01:09.720 --> 01:10.200 stuck. 01:10.480 --> 01:14.000 And you can improve that by adding more agents to the mix. 01:14.000 --> 01:19.090 For example, you could have a planning agent that first of all, that's the first. 01:19.130 --> 01:24.770 The main assistant is also a planner and decides on what tasks need to be done, and then delegates 01:24.770 --> 01:27.210 them to the execution to the worker. 01:27.450 --> 01:29.250 Um, and so you could do it that way. 01:29.370 --> 01:31.130 These things all have pros and cons. 01:31.170 --> 01:35.850 The benefit is that you've then you're able to divide a bigger problem into smaller steps, and the 01:35.850 --> 01:38.090 autonomous agent can figure out how to do that. 01:38.290 --> 01:43.890 The downside is that it makes it harder for the agents to kind of plot its own course. 01:43.930 --> 01:49.090 And based on what comes out of that first task, maybe it would decide to do things a bit differently 01:49.090 --> 01:49.930 after that. 01:49.930 --> 01:55.610 So by by planning you sort of you lose some of that, that freedom that you get from doing it all with 01:55.610 --> 01:56.290 one agent. 01:56.450 --> 01:58.730 So these are all things to experiment with. 01:58.770 --> 02:00.250 There's no right answer. 02:00.290 --> 02:01.650 More than anything else. 02:01.650 --> 02:04.730 You need to be willing and ready to experiment. 02:04.730 --> 02:08.650 But add on tools, add on a sequel tool that's easy to do. 02:08.810 --> 02:12.210 And on the topic of sequel, you should also improve the memory. 02:12.210 --> 02:17.970 The memory we're using is just the basic in-memory memory, and you should change that to be the sequel 02:17.970 --> 02:22.610 memory that we already did earlier in the week, and just put that in there, and that way it will be 02:22.650 --> 02:28.530 able to remember who you are if you come back, if you use radios, uh, feature to be able to log in 02:28.530 --> 02:34.340 and identify yourself, then you could use your username as the conversation thread and be able to continue 02:34.340 --> 02:42.220 a conversation and keep tasks build up a sort of library of tasks that you have so, so much you could 02:42.220 --> 02:42.820 do with this. 02:42.860 --> 02:49.260 It's a it's like I'm, I'm rather than giving you specific instructions, I'm asking you to be autonomous 02:49.260 --> 02:49.860 with this. 02:50.380 --> 02:55.700 Go and show your own agency and build this into something amazing, absolutely amazing. 02:55.700 --> 02:56.820 And then share it. 02:57.140 --> 03:02.140 Post about it on LinkedIn so I can weigh in and see examples and and show me the tools. 03:02.140 --> 03:07.260 Put them in community contributions so that people can add them in, put in your tools, libraries that 03:07.260 --> 03:10.660 people could import and other functionality. 03:10.820 --> 03:14.460 And I just can't wait to see where people take this project. 03:14.460 --> 03:18.620 And I hope, like me, that you're something of a convert at this point. 03:18.620 --> 03:24.060 Maybe you always did like Landgraf, but but now, uh, now I'm drinking the Kool-Aid. 03:24.100 --> 03:26.780 I'm really, really happy about this week. 03:26.780 --> 03:28.300 It's been very exciting. 03:28.700 --> 03:31.460 And with that, it feels like we were only just starting week four. 03:31.500 --> 03:38.060 And now we're saying goodbye to week four, goodbye to Landgraf, and hello to week five as we move 03:38.060 --> 03:40.940 on to Autogen from Microsoft. 03:40.940 --> 03:43.860 And there's a lot to show you I can't wait. 03:44.260 --> 03:45.940 See you for week five.