Learning Library

← Back to Library

LLM Fluency Scale Explained

Key Points

  • The video introduces a model‑agnostic “LLM fluency scale” to help users gauge their AI proficiency, noting that most people fall below level 5.
  • Level 1 (basic beginner) covers typical users who employ tools like ChatGPT or Copilot for simple tasks such as rewriting emails or editing documents.
  • Levels 3‑5 focus on developing a mental model of how large language models work—understanding token prediction, reasoning limits, and the importance of context retrieval.
  • The presenter offers practical resources, including a comprehensive assessment prompt and a 90‑day custom development plan, to guide individuals from their current level toward their personal AI goals.

Full Transcript

# LLM Fluency Scale Explained **Source:** [https://www.youtube.com/watch?v=DdlMoRSojtE](https://www.youtube.com/watch?v=DdlMoRSojtE) **Duration:** 00:14:28 ## Summary - The video introduces a model‑agnostic “LLM fluency scale” to help users gauge their AI proficiency, noting that most people fall below level 5. - Level 1 (basic beginner) covers typical users who employ tools like ChatGPT or Copilot for simple tasks such as rewriting emails or editing documents. - Levels 3‑5 focus on developing a mental model of how large language models work—understanding token prediction, reasoning limits, and the importance of context retrieval. - The presenter offers practical resources, including a comprehensive assessment prompt and a 90‑day custom development plan, to guide individuals from their current level toward their personal AI goals. ## Sections - [00:00:00](https://www.youtube.com/watch?v=DdlMoRSojtE&t=0s) **Model‑Agnostic AI Fluency Scale** - The speaker introduces a model‑independent framework that grades users' LLM proficiency on a ten‑point scale and provides assessment prompts and a 90‑day development plan to guide improvement. - [00:03:19](https://www.youtube.com/watch?v=DdlMoRSojtE&t=199s) **Thinking Backwards for Prompt Engineering** - The speaker explains that grasping AI mental models lets users focus on the desired output instead of the prompt itself, fostering intuitive, outcome‑driven prompt engineering across beginner to advanced skill levels. - [00:07:57](https://www.youtube.com/watch?v=DdlMoRSojtE&t=477s) **Teaching as a Learning Catalyst** - The speaker emphasizes that teaching forces clarity, reveals knowledge gaps, and by documenting and sharing AI insights—through curricula, workshops, or media—one both scales influence and deepens personal understanding. - [00:11:48](https://www.youtube.com/watch?v=DdlMoRSojtE&t=708s) **Mapping Career Fluency to AI Agents** - The speaker urges professionals to treat AI skill development as a fast‑moving train, build mental models of career fluency levels, and proactively map emerging AI‑agent competencies onto those levels to stay aligned with future demands. ## Full Transcript
0:00You know, one of the questions that I 0:01get the most is, "How do I level up an 0:03in AI?" And how do I know that I'm doing 0:06good or at least improving? There hasn't 0:08really been a comprehensive approach 0:11that is agnostic of models, that doesn't 0:13care if you're a Chad GPT user or a 0:16Copilot user or a Claude user. It just 0:19focuses on the principles and your level 0:21of understanding and helps you to scale 0:22it up. That's what I'm doing here. In 0:25this video, I'm going to walk you 0:26through how to tell roughly where you 0:29are in the overall LLM fluency scale. 0:32Spoiler alert, most people end up below 0:35five. This is a tough scale. Don't be 0:37afraid. Then I'm going to walk you 0:39through what it looks like to actually 0:42improve. Now, you're going to have a lot 0:44more to dig into here. In the post, I 0:46have a comprehensive assessment prompt. 0:48I also have a prompt for building 0:49yourself a 90-day custom development 0:51plan for wherever you're at. So, there's 0:53lots to dig into, but first, let's 0:56understand the levels and what they 0:58mean, and I'm not going to do all 10 1:00because people just don't stick around 1:01for that. I'm going to give you some 1:03blocks, right? Like a one to a three, a 1:05three to a five, that kind of thing. And 1:06we're going to go pretty quick here. So, 1:08number one, level one, basic beginner 1:10level. Most people are here. That's your 1:14default. If you are a chat GPT user, a 1:16co-pilot user, if you're the kind of 1:18person who uses these AI tools to 1:21rewrite emails, to adjust a document, 1:24you're probably in this one to three 1:26area. And I just want to emphasize 1:28again, this is not a bad or good thing. 1:30This is just helping you understand 1:31where you are so that you can figure out 1:33where you want to go and where your 1:35goals are. Not everyone has to be a 10, 1:37right? Like that's not the point. The 1:38point is to understand your level and 1:40what your goals are and make sure you 1:42are equipped to get there. And that's 1:43what I'm all about. Let's jump ahead to 1:45number three to five here. What does 1:47that look like? So 3 to five, you are 1:49starting to build a mental model for AI. 1:52And this is why this scale is so 1:53important. By the way, no one talks like 1:55this. Like people tend to give you 1:57specific skill sets and and I can do 1:59that, too. I'm going to talk about some 2:00of the specific skill sets you 2:02demonstrate, but you need an overarching 2:04perspective on the fluency and 2:06competency assessment that you're 2:08looking for at this level. And 3 to 5 is 2:10all about building mental models. You 2:12are starting to understand how LLMs 2:14actually work, what they do when they 2:17reason, what they do when they don't 2:18reason. You're starting to understand 2:20that LLM don't truly know things, that 2:22they're not programmed. You're 2:24understanding what next token prediction 2:26looks like. You have the beginnings of a 2:28mental model of what AI can do. Now, one 2:32of the things that is more important 2:33these days than it used to be is 2:36understanding context retrieval. It used 2:39to be that if I gave you those like 2:41understand how LLMs work lessons, that 2:43would be enough. But now, as AI has 2:45gotten more powerful, you actually do 2:48need to understand the ability to 2:51retrieve a larger piece of context and 2:53work with it. Because to be honest, 2:54these AIs can take booksized prompts 2:57now, right? Book-sized context windows. 2:59And so, you need to understand how that 3:01works a little bit and have a mental 3:02model for that, too. I hasten to add, 3:05none of this means you can build an AI 3:07system. None of this means that you can 3:09build a context window like a rag system 3:11or a memory system. If that's all above 3:13your head, you're still firmly at 3 to 3:15five if you have the mental model down. 3:17The last piece I want to call out from a 3:19mental model perspective is that this 3:21conceptual understanding of AI is going 3:24to naturally lead to you thinking 3:26backwards from outcomes. You're going to 3:28stop asking what should I tell the AI 3:31and at this stage you're going to start 3:32asking what is the output that I need? 3:35because the mental models are going to 3:37inform your understanding of how it 3:39creates the outputs and you're naturally 3:40going to start to say, "Okay, I get a 3:41sense of how the sausage is made, 3:43right?" And so this is the output I 3:45want. I can work back in my head. And 3:47this is how you start to get to what I 3:49would call intuitive prompt engineering. 3:51You're not reading from a book. You're 3:53not trying to copy a prompt necessarily. 3:55Maybe you do, maybe you don't sometimes. 3:57And even if you do, you know how to 3:58massage it and tailor it a little bit. 4:00Or you can write it yourself, but you 4:02know how to get to the outcome you want. 4:03So many people are here. I would say 4:05what I just described with sort of the 4:06level one and two where you're just 4:08basic users of co-pilot or a basic user 4:10of shed GPT plus this level with 4:12understanding LLM and how they work with 4:15mental models and kind of going from 4:16there. That's almost almost everybody 4:18right like if you want to talk about 4:208020 80% are right there. Now what goes 4:24on above those levels? I'm going to make 4:26this as accessible as possible. And I'm 4:27going to give you a sense of whether you 4:30need to go farther or not based on your 4:32goals. So from 5 to 7, you really are 4:35probably going to be working with AI on 4:38a professional basis very seriously. So 4:41if you get above five, if you get above 4:42this mental model session, there are 4:45some patterns that start to come through 4:48that you just don't see at a lower 4:50fluency level. And I'm going to name a 4:52few of them, but you're going to get the 4:53idea. The overall approach is systemat 4:55system systematization. You are using 4:57systems thinking moving from a five to a 5:00seven. And that applies to AI because 5:02you take it very seriously. So a person 5:04between a five and a seven is going to 5:06be thinking in auditable patterns with 5:08AI. They're going to be thinking in 5:10terms of usually do this and they're 5:12going to move that over to this is the 5:14sequence I follow. I get a predictable 5:16result. I know how to get the 5:17predictable result and I can start to 5:19systematize it in a way that others can 5:20do it too. You see that difference? It's 5:22not just an intuition at that point. 5:24It's actually a understanding of how the 5:26system works so that you can predict and 5:28move with it. Another example of 5:30systematic thinking is building for 5:33prompt yield. So prompt yield is like 5:35what is your quality output per unit of 5:37prompting. So if you're prompting 5:40inefficiently, you might take 10 5:41iterations to get one usable output. But 5:44if you're prompting efficiently, you 5:45might do one or two prompts and get 98% 5:48of the way there and move on. And that 5:50is part of why, by the way, I emphasize 5:52the kinds of prompts I emphasize in my 5:55posts. I think it's really important to 5:57value the tokens, value the time that we 5:59are taking with AI so that we can go on 6:01to other things. And it is much much 6:03much more efficient to just do the 6:06prompt correctly and just get the right 6:08answer. And someone who is building and 6:10thinking in systems is able to 6:12understand that and also able to move 6:14from a casual intuitive I think this is 6:16the right prompt to get to this output 6:18to a systematic this is the yield I get 6:21on this prompt. I think this prompt can 6:23be modified in these three ways and I'm 6:25going to get a much more efficient 6:26output and then they make the change and 6:27they measure it and they see it. These 6:28kinds of people think in feedback loops, 6:30right? Like you're thinking in terms of 6:32my systems working to make me more 6:35effective at AI. doesn't necessarily 6:37mean that you have to have tons of 6:39tools, but in my experience, most people 6:42at this stage will have a prompt 6:43library. They will have five to seven 6:45tools they're working with regularly in 6:48the AI space. They will have preferences 6:50for specific work tasks that are 6:53associated with those tools and they 6:55will be seen by their teammates as a 6:57peer collaborator and peer leader who 7:00can help the team put in place systems 7:02that matter. So far so good. You notice 7:04by the way these are not job specific. I 7:08am not giving you the fluency levels for 7:10engineers and then the fluency levels 7:12for for PMS. Do you know why? It's 7:15because I have a strong conviction that 7:17AI is a generalist skill set and we are 7:20probably teaching it wrong if we dive 7:23too deep into verticals without that 7:25generalist conceptual foundation. And we 7:28really haven't had that. And that's what 7:29I'm setting out to do here. I think it's 7:31great if you understand how to build 7:33with Langsmith as a developer, but I 7:35don't think that's the only kind of AI 7:38learning and grounding you need. And I 7:40think that we're missing this piece 7:41here. This sort of general approach to 7:43skill sets and fluency. And I think 7:45having a common understanding here will 7:47be helpful. Let's jump to 7 to 9. What 7:50does it look like? Really, at this 7:52point, you've mastered systems thinking. 7:54You understand how LLMs work. You are a 7:57teacher and you are a trailblazer. And 8:00so you need to start thinking about who 8:02you can teach with your skill set and 8:05how your teaching drives your own 8:08learning. So I will say for me teaching 8:10has been super helpful in driving 8:12clarity and revealing gaps in my own own 8:15understanding that I have to 8:16relentlessly close. Most teachers will 8:18tell you regardless of subject that 8:20that's true. Try to be if you're at this 8:23level a documentarian. And what I mean 8:26by that is the more you document about 8:29what you're learning and what you're 8:30thinking and how you're growing, the 8:32more you're able to scale your influence 8:35and teach others. And it's not about 8:37growing influence. It's about being able 8:39to communicate really clearly things 8:41that are net new in the space that you 8:44can then understand how to teach others 8:47in a way that is accessible for their 8:49level. And so that might look for you 8:52like setting up the AI training 8:54curriculum at work. It might look for 8:56you like leading a group of developers 8:58through through their first AI build. It 9:01might look for you kind of like what I 9:03do here where you're on YouTube or 9:05you're on Substack and you're kind of 9:06like talking through what it means to 9:08grow and learn AI. There are lots of 9:12ways to do this, but the systematic 9:15thinking doesn't go away. And so you're 9:17not just thinking in team systems or 9:18personal systems. You are often doing 9:20something that is public that many, many 9:23others can use. So you might be building 9:24a clawed projects that others can use. 9:27You might be building a little vibe 9:29prompted tool that others can use in 9:32order to understand their own level of 9:34fluency. Similar to what I've done with 9:36the prompts in this in this piece 9:38actually. But your goal is to pull the 9:42impossible problems into the realm of 9:44the possible. That's what that 9:46innovation piece looks like. someone who 9:48is teaching, who is learning, who is 9:50growing, they should be helping to pull 9:52forward things that were previously 9:54deemed very difficult to do with AI 9:57because they are helping to discover AI 9:59capabilities. And by the way, that 10:01understanding that AI capabilities are 10:03not all documented and you can discover 10:06them and you can put them to new uses is 10:10a great example of what teaching and 10:13innovation and the relationship between 10:15systems thinking and deep understanding 10:17of LLMs is all about. People who 10:20understand LLM deeply know that LLMs are 10:24not all discovered. We do not ship an AI 10:27and OpenAI knows all about it. We ship 10:30an AI and we all collectively discover 10:33the capabilities it has because it's 10:35more accurate to say these systems are 10:37grown than to say that they are 10:39programmed. And so we're all discovering 10:41together what grew. And that is part of 10:43the job at level seven to nine is to 10:45start to innovate and understand where 10:48to push farther on LLM capability and 10:52why it matters and then be able to turn 10:53around and teach that back and really 10:55grow the practice. So 10:58I don't give tens. There's not going to 11:00be a 10 here. I think one of the things 11:02that I want to call out is that you 11:03should understand 11:06that wherever you are, your competitive 11:10reality is shifting. We are in October 11:14of 2025. We are not too far away from 11:18the end of the year in 2026. You need to 11:20think of your baseline as shifting into 11:24the new year such that like the whole 11:27population is going to grow into one to 11:30three in the next year and there will be 11:32a much larger part of the population 11:34growing into that sort of 3 to five area 11:38and there will be many more people who 11:39are pushing themselves up the skill 11:40ladder from there. What I'm saying isn't 11:43here to sort of panic you. Your goal may 11:45not be to be a teacher or an instructor. 11:46Maybe your goal is to be a systems 11:48thinker. I don't know. or maybe you're 11:49perfectly happy just understanding how 11:51LLMs work. But regardless, I want you to 11:54recognize that the skills required at 11:57each stage are sort of evolving as you 11:59go. My best advice, and I think I've 12:01said this other times, is think of it as 12:03a moving train and it's never going to 12:05go slower than it's going right now. So, 12:07hop aboard and get yourself going as 12:09quick as you can in a way that you feel 12:11comfortable that feels aligned with your 12:13goals. And so just as I said that three 12:15to five sort of people learning systems 12:17thinking and all of that uh are are 12:20starting to develop mental models of AI, 12:22develop a mental model of your career 12:25path a little bit. Have a sense in your 12:27job family of what is the level of 12:29fluency that would be useful. And then 12:32here's the extra credit. I want to give 12:33this to you because come back to 2026. 12:36You're going to want this. Think about 12:38the corresponding skill sets that will 12:41emerge and map them onto this fluency 12:44chart. Let me give you an example of 12:45that. That can feel really abstract. 12:47Think about AI agents. We just had on 12:50October 6th a launch of a new kind of 12:53agent framework from OpenAI. Well, think 12:56about what the fluency types map onto 12:58that, right? Like how do those map onto 13:00it? Well, systems thinkers are going to 13:01think about how you build not just one 13:04agent but multiple agents, how you 13:05sustain them within an org. Innovators 13:07are going to think about new things you 13:09can do with agents. People just 13:11understanding LLMs are going to think 13:12about what is an intuitive way to get a 13:14task done that helps me to express my 13:17understanding of LLMs and get real work 13:19accomplished. And people who are just 13:21starting out are going to scratch their 13:22heads and say this this agent thing 13:24looks really hard. But you can you can 13:26map that whole technical launch onto 13:28this capability assessment and you can 13:30do that with other launches that are 13:32coming forward too. And so this is not 13:34meant to be an October 2025 artifact and 13:37we're done. It is actually meant to be a 13:41living breathing framework that helps 13:44you make sense of your own skill level 13:46relative to where we are with AI that 13:48you can come back to again and again. So 13:50there you go. That is my evergreen AI 13:53fluency assessment. And as far as I 13:54know, we haven't really talked about 13:56stuff like this before, or certainly not 13:58in this way. I hope you enjoyed it. I 14:00hope it's helpful, and I'd love to hear 14:02where you're at, and most of all, where 14:04you want to get to. That's one of the 14:06things I was really excited about for 14:08this particular piece is I wanted to put 14:10together a sense of the ladder, for lack 14:14of a better term. It's really not a 14:16ladder, but the sense of the jungle gem 14:17of AI and a sense of where people can 14:20go. And then here where all of you want 14:21to go and start to craft some prompts 14:23that help out and all of that. So yeah, 14:25drop a note.