WEBVTT 00:00.160 --> 00:10.920 I explained to you the difference between a assistant rack agent and multi-agent systems. 00:11.680 --> 00:14.520 Clients often ask me to clarify these terms. 00:14.720 --> 00:16.400 So let me break it down for you. 00:16.400 --> 00:24.720 Understanding these concepts will save you a lot of time when it comes to building, using, or selling 00:24.760 --> 00:26.280 these AI systems. 00:27.800 --> 00:30.960 I will be discussing it using practical examples. 00:31.000 --> 00:32.560 Let's start with the basics. 00:33.120 --> 00:34.160 Assistance. 00:34.600 --> 00:41.680 So the best example of an AI assistant is ChatGPT, which is essentially a chatbot powered by large 00:41.680 --> 00:46.040 language models that are trained on massive amounts of data. 00:46.360 --> 00:51.600 They are designed to help you with specific tasks, mostly by answering your questions. 00:52.040 --> 00:58.640 And they are trained on general knowledge, but they can't provide answers about your internal data 00:58.840 --> 01:00.520 unless you fine tune them. 01:01.080 --> 01:07.680 So they are great for answering questions and generating text, but are limited as their knowledge is 01:07.680 --> 01:12.720 cut off if they are not connected to the internet or external data sources. 01:13.640 --> 01:17.880 They are designed for direct interaction through voice or text. 01:17.880 --> 01:21.480 Interactions with AI assistants are straightforward. 01:21.520 --> 01:24.080 You provide input and they give you an output. 01:24.920 --> 01:29.600 They rely solely on the llms training data, which is often outdated. 01:29.840 --> 01:34.880 Now let's move on to RAC, which stands for Retrieval Augmented generation. 01:35.960 --> 01:40.960 So this is a fancy way of saying let's give our access to extra knowledge. 01:41.840 --> 01:49.320 So while AI systems like ChatGPT rely on the training data, RAC systems or by default connected to 01:49.360 --> 01:51.240 external knowledge sources. 01:51.720 --> 01:58.930 So this means they can pull in real time or context specific information that's not part of their original 01:58.970 --> 01:59.610 training. 02:00.290 --> 02:04.050 For example, a chatbot on a company's website. 02:04.690 --> 02:11.850 Instead of only answering generic questions like ChatGPT, a rack system can access the company's frequently 02:11.850 --> 02:12.730 asked questions. 02:13.210 --> 02:20.450 Products or services descriptions, for example, stored in Google Docs or vector databases to make 02:20.450 --> 02:23.290 it easy for large language models to fetch. 02:24.450 --> 02:32.490 In short, rack enhances the AI assistants ability, like chatbots, to give you more relevant responses 02:32.650 --> 02:38.290 by combining its language generation skills with external knowledge retrieval. 02:39.010 --> 02:48.290 Next, agents to AI agents are like AI assistants or rack, but they don't just access external knowledge, 02:48.530 --> 02:53.090 they also connect to external tools and know how to use them. 02:53.450 --> 03:02.300 Thanks to the prompts you define so these systems are autonomous so they can take action on your behalf 03:03.300 --> 03:04.780 without constant input. 03:05.420 --> 03:15.420 For example, agents can integrate with tools like CRMs, Gmail and Google Calendar to automate tasks. 03:15.860 --> 03:21.060 And they might draft an email, send it scheduled meetings, or even update your CRM. 03:22.580 --> 03:31.980 While they can access external knowledge like RAC and have access to the internet or a company knowledge 03:31.980 --> 03:35.900 base, they don't need to be called agents. 03:36.020 --> 03:38.420 The key distinction is their ability to act. 03:38.860 --> 03:41.820 So for example, they don't just draft an email. 03:42.020 --> 03:45.220 They will draft it, send it and confirm it's done. 03:45.620 --> 03:52.340 So agents are about action and taking the next step, not just answering questions. 03:52.660 --> 03:54.980 Now let's talk about multi-agent systems. 03:55.740 --> 04:01.380 Why Wide area assistance systems and agents each work independently. 04:01.620 --> 04:09.060 Multi-agent systems combine multiple agents, each with a specific role, into a collaborative team. 04:10.100 --> 04:16.500 Think of it as having a group of specialists who coordinate to handle complex, complex tasks. 04:17.220 --> 04:25.860 So a supervisor agent acts as a manager, delegating tasks to the right agent. 04:27.500 --> 04:31.220 So the calendar agent to schedule meetings. 04:31.700 --> 04:34.900 Email agent to drafts and send emails. 04:35.380 --> 04:43.820 And the research agent gathers information from the internet or other database or the project agent 04:44.580 --> 04:47.100 which updates workflows or project data. 04:48.260 --> 04:54.820 So imagine you message the supervisor agent, schedule a meeting with Sara. 04:55.230 --> 05:00.550 You may have heard the agenda and find the latest AI trends for our discussion. 05:01.230 --> 05:07.190 So the supervisor agent reads your request and splits it into tasks. 05:08.270 --> 05:11.430 So the calendar agent schedules the meeting. 05:11.910 --> 05:15.990 The email agent drafts and sends the agenda to Sara. 05:16.150 --> 05:20.790 The research agent gathers the latest trends. 05:22.150 --> 05:28.030 All these agents send the results back to the supervisor agent. 05:28.510 --> 05:33.350 And the supervisor agent consolidates the results and reports back to you. 05:33.710 --> 05:40.310 So basically each agent specializes in one task, but together they achieve more. 05:40.470 --> 05:45.510 So tasks are done simultaneously and it saves a lot of time. 05:45.830 --> 05:48.070 You can add more agents as needed. 05:48.630 --> 05:53.150 And these systems can handle more dynamic and complex requests. 05:53.790 --> 06:01.110 So in short, multi-agent systems are like a team of autonomous experts led by a supervisor. 06:01.670 --> 06:08.950 The key difference between multi-agent systems and others we have discussed is this collaborative aspect. 06:09.830 --> 06:14.390 Because AI assistants work alone. 06:16.550 --> 06:24.470 Rack systems have access to external sources of data and enhanced knowledge retrieval. 06:27.110 --> 06:32.630 Single agent can use tools but work independently. 06:33.150 --> 06:41.910 And multi-agent systems bring multiple specialized agents together to solve problems that might be too 06:41.910 --> 06:44.750 complex for a single AI agent. 06:44.990 --> 06:49.150 So to conclude, the key differences are independence. 06:49.470 --> 06:57.390 So agents and multi-agent systems are highly independent AI assistants need user commands. 06:57.910 --> 07:06.670 RAC are not autonomous and they enhances large language models collaboration. 07:07.630 --> 07:10.270 So multi-agent systems excel at teamwork. 07:11.350 --> 07:17.550 Agents usually work alone and assistants and RAC don't collaborate at all. 07:19.350 --> 07:20.950 Third knowledge integration. 07:21.270 --> 07:27.350 RAC specializes in adding external information and others rely on pre-existing knowledge. 07:27.870 --> 07:29.790 The next one tax complexity. 07:30.030 --> 07:35.710 Multi-agent systems handle complex, more sophisticated distributed tasks. 07:35.910 --> 07:38.710 And the last one adaptability. 07:39.350 --> 07:47.190 Multi-agent systems and agents can adapt to changes a assistance just follow fixed rules. 07:47.390 --> 07:52.350 And RAC retrieves new info but doesn't learn or evolve.