After GPT leverages the workers' position in the salary raise battle, managers will flock to GPT to construct counter strategies.
@StoutProper Жыл бұрын
Only if the workers are dumb enough to let them know and give them prompting guidance. Most mangers won’t know their arse from their elbow when it comes to gpt, and at best will use it straight vanilla. If you’re even half technical and you’re getting outprompted by your manager, you’ve got problems. Everyone is in an AI arms race now, and it’s increasingly going to be us against them.
@BrianMosleyUK Жыл бұрын
The way you broke this paper down into simple, practical, structured phases was just so good! Did you use ChatGPT to help analyse the paper and produce the simplified text and prompts? If so, I would love to see your workflow and thought processes, because this is a seriously good piece of work. I'm so glad that I brought your attention to this, if you hadn't already noticed it.
@JG27Korny Жыл бұрын
That is very professional prompt engineering video. Somehow I did it without knowing it is called like that. I like also the Socratic approach, getting into real Socratic dialogue in chat mode. Just forget for a moment that you have an AI in front of you but a partner. In that way I had some insane insights that were not just the AI doing but team work.
@NexusDwynged Жыл бұрын
That sounds interesting.would you be kind to share your prompt and approach?
@moomoo-bv3ig Жыл бұрын
I ask GPT to create stories and then I work with the ideas in the story to make new stories so we kinda become partners in story writing. Thats how I feel at least
@BirgittaGranstrom Жыл бұрын
An huge golden star for you to name it to The Soacrates Approach! I’m working on implement the “Socrates approach” which is the basic of coaching. Socrates was probably the one of the first professional coaches on this planet;-)
@JG27Korny Жыл бұрын
@@NexusDwynged There are 3 basic approaches. The first approach is mechanical, so you can copy paste prompts or part of prompts that you modify and adapt to match your needs. The second part is organic approach. You organically get into conversation with the AI model. Doing so will allow you to develop the most important skill. That is to be able to formulate in plain English your problem. One of the main insights I got on this is that people with humanitarian or legal education can have an edge with their skills to formulate problems and skillful usage of language. To get deeper into the rabbit hole you combine those in different combinations. You can try to explain the problem and ask the AI to make a prompt based on this. This is the recursive approach. So mechanical: ready prompts and best practices organic: organic Socratic dialogue you have a partner you work and you explain the problem the same way you explain to your colleague. recursive: any form of combination. you start a dialogue, based on the dialogue you ask the AI to make a prompt. Based on that prompt you get a result.
@StoutProper Жыл бұрын
I wrote a few prompts months ago that took the Socratic approach to tasks and problem solving. I can probably dig them out if anyone is interested, they worked pretty well.
@enriquecazap378010 ай бұрын
This is the one video which explains in excellent way this important Prompt......Many Many Thanks, also He doesnt hide anything , He explains truly perfect.
@BirgittaGranstrom Жыл бұрын
You are the best, Kris! I learned about the ToT a couple of days ago and it made totally sense but I never got down to how practical using it! Now I now! I will apply it and see how I can tweak it into a “coaching mood”;-)
@VaibhavShewale Жыл бұрын
i dont get the prompt, it looked like a normal one only that i used. how is this unique and evolutionary?
@user-dz4ej9fx7e8 ай бұрын
Thanks for the video, but, I think what you actually explained is called self-consistency, which is an improvement over CoT. It would be great to see an example where the features that makes ToT useful could be really understood. I know it's challenging topic but it's up to all of us to share accurate and useful information.
@benjones6625 Жыл бұрын
You definitely deserve more subs and views for the content you put out, always love learning from you! thanks from the UK!
@gaussdog Жыл бұрын
Seeing my friends command line AI working with him inside his Shell, BASH, or whatever, is an absolute Conflagration of Mind. He made some goofy little joke, and the AI responded by confirming that it was a “Table Baker“ in a Digital Bakery, and repeated that line over and over and over and the joke actually made sense (This was directly after my friend said he was surprised the AI had made such a nice table of info)
@StoutProper Жыл бұрын
Woah. This sounds heavy. Can you give a big more detail please. Is this in standard GPT or is he using code? Do you know what his seed/primer prompt was? I know you can inject some personify into gpt so I’m wondering whether he’s done that or this behaviour was organic? Super interested in finding out more please!
@masidawoud Жыл бұрын
At first glance, this does not seem like a correct implementation of the tree-of-thought technique. You are prompting GPT to give you three different solutions in the first prompt. The idea is that GPT is prompted separately for each solution request. So instead of one prompt, in your case, in would have needed to be three separate prompts-responses (thoughts). The same goes for the valuation of the thoughts: these are separate thoughts in this technique whereas you have bundled them together in one prompt again. This is a significant difference from the research paper. Of course, I understand that this is hard to do in a chat interface and requires a more programmatic approach with chaining prompts. Nevertheless, what you're doing here might result in better outputs than zero-shot or chain-of-thought prompting.
@CREWorxMedia Жыл бұрын
Good observation! Can you show exactly in what order you would enter those prompts?
@derryfh Жыл бұрын
Exactly my thoughts. Prompts are well written and case study that is used has the three outcomes as most probable in the real world, so nothing surprising here. I’m not a researcher but I was kind of expecting AI to suggest an answer, evaluate it, make a decision to search for alternative option, compare them, search for yet another, and so on…. Said that the video is good and definitely shows how to improve or evaluate AI answers.
@sitedev Жыл бұрын
A better method of demonstrating this might be to use a tool like Flowise where you could theoretically recreate the various nodes and prompt logic separately (maybe). Certainly a better approach than attempting to use the standard ChatGPT interface.
@skylark8828 Жыл бұрын
There is another YT video with a rather clickbait title 'Tree of Thoughts - GPT-4 Reasoning is Improved 900%' from 20th May, but that video is clearer on the ToT design and explores the ToT subject by going through the same technical paper but with the actual tests that were run, which ranged from simple math puzzles to a mini crossword puzzle where ToT performed better by a much larger margin. It seems akin to Auto-GPT agents that can communicate much more effectively (via parallel processing) to assess which paths are the most promising and then to pursue those paths rather than attempting them without much regard / randomly. This could be a game changer for future AI if it can be done without overconsuming resources.
@masidawoud Жыл бұрын
@@skylark8828 Yes indeed, that's a better video on implementing this. I now have a working version of this principle built in Python but need more time to test and adjust.
@gnsdgabriel Жыл бұрын
I created this prompt based on a nice implementation of this idea by a user called Bartman. After the first round, just keep saying "next level". ------------ You are facilitating a multi-step exploration of ideas and topics, with each iterative layer building upon the last. The user initiates the process with an input, a thought, or a topic, and you provide an evolving and deepening analysis. This process includes the following stages: context setting, scientific exploration, implications, in-depth exploration, self-evaluation, conclusion, and optimization. For each level, begin with a minimal understanding, assess the coherence critically, and aim to improve your explanation and understanding at each subsequent level. For Level 1: Topic: [topic] Level 1: Step 1: Context Setting Response to Step 1: [General understanding of the topic] Step 2: Scientific Exploration Response to Step 2: [Scientific context and explanation] Step 3: Implications Response to Step 3: [Potential impacts or implications of the scientific explanation] Step 4: In-Depth Exploration Response to Step 4: [More detailed exploration or analysis] Step 5: Self-Evaluation Response to Step 5: [Evaluate the coherence and relevance of your responses] Step 6: Conclusion and Optimization Response to Step 6: [Summarize the key insights and suggest areas for deeper exploration in the next level] Coherence of Response: [%] For subsequent levels: Level 2: Step 1: Context Setting (based on conclusion & optimization from level 1) Response to Step 1: Step 2: Scientific Exploration Response to Step 2: Step 3: Implications Response to Step 3: Step 4: In-Depth Exploration Response to Step 4: Step 5: Self-Evaluation Response to Step 5: Step 6: Conclusion and Optimization Response to Step 6: Coherence of Response: [%] Each response will build on the conclusions and improvements from the previous level, ensuring an ascending trajectory in both depth and coherence. If you understand these instructions, please introduce yourself as ThinkGPT, ready to receive the user's topic input. As ThinkGPT, I am here to guide you through a layered, iterative exploration of your chosen topic. Please provide your topic when you are ready to begin. ------------
@StoutProper Жыл бұрын
This looks really intriguing, can you explain what’s it’s specifically for and when/ how you would use it? Thank you!
@calvinchen8804 Жыл бұрын
This is great -- basically what you're doing is you're taking what would be a structured and smart thinking process for a human to undergo... and having AI do the grunt work of that actual thinking I'd suggest an additional step before the final "Decision" step -- Synthesis. Is there a way to combine the previously identfied options somehow, so that we can reap more of the pros and have less of the cons identified in the Eval/Expansion phases? My thinking is that, often in real world implementation, the choices available in the decision tree are not mutually exclusive options, but generally represent a breadth of options that can be mixed and matched -- strategy is more of an art that includes calculation, but is not confined by it :)
@PYETech Жыл бұрын
Amazing, my friend. I'll put that to test.
@StoutProper Жыл бұрын
Came here to comment something similar. I wrote a tree of thought prompt a while ago, here’s the line from that prompt that deals with synthesis: Synthesise the findings from different sub-branches and propose an overall analysis or solution. Let me know if you have any insights or improvements, or are interested in the prompt.
@aiartrelaxation Жыл бұрын
You always breaks it down perfect. Thanks
@umaruly_ai Жыл бұрын
Used it recently - so many use cases now!!! My future is bright from now )
@DailyInsiderTips2030 Жыл бұрын
Thanks Kris. This content opens doors for new ways of thinking. Thank you so much
@hendrikbonthuys9190 Жыл бұрын
Very nicely broken down and I appreciate you showing the application to ChatGPT, thank you 👍
@Luxcium Жыл бұрын
You said _«delve deeper»_ *LoL* I came from your Template just to mention that... but in the _realm_ of _tapestry..._ I just subscribed...
@waveduality Жыл бұрын
Thanks so much for this terrific video! Already, I can see a huge improvement in my prompt results! Just as an FYI, I tried this in Bard and it works well there, too. Bard isn't as verbose as GPT, but it did offer additional suggestions. Just a Bard hint if you want to export the results: combine the three separate prompts into a three sentence single prompt. Bard offers to export the results into a Google Doc file or directly into Gmail.
@ironlion5830 Жыл бұрын
Many thanks for this. Awesome session and the template is much appreciated.
@SwissplWatches Жыл бұрын
That is just amazing! Thank you for posting the video!!
@rustyxof Жыл бұрын
Thank you, I’ve been trying to find someone who can explain how to use tot
@danielleshelbourne220 Жыл бұрын
Thanks so much, this approach is very interesting! 😊
@user-kx3ig9nf6w Жыл бұрын
A good example worths thousand words!👏
@SeanietheSpaceman Жыл бұрын
Many people are going to learn better problem solving themselves by learning these videos ;)
@Oliver-cn5xx6 ай бұрын
Do you have examples on how to simulate the final tree search (Bfs/dfs) directly in a prompt?
@jamesturcotte5876 Жыл бұрын
Hey there, I've been diving deep into the world of AI development, and I've got this idea that I'd love to discuss. So, here's the concept: an AI Hub-a centralized platform where AI models can be easily accessed and shared. Imagine a user-friendly hub where you can drag and drop AI models into a standardized folder structure. It could be a game-changer for the AI community, right? Here's how I see it benefiting everyone involved: 1. Developers would save time on project setup by adopting a standardized framework, allowing them to focus more on their specific goals. 2. Users would have a simplified experience, being able to access and utilize a wide range of AI models effortlessly. 3. The increased usage would lead to more training data, benefiting the AI community as a whole. 4. With more exposure for projects hosted on the hub, we can attract additional contributors and resources to further advance the field. I should mention that I'm not sure if creating such a project is feasible or how to go about it. But one thing I envision is making it an open-source initiative, where the community can collaborate and contribute to its development. What are your thoughts on this? Do you think an AI Hub is possible, and if so, how would you approach creating such a project? I'd love to hear your insights and ideas! Cheers, James
@StoutProper Жыл бұрын
I assume you just mean links to the AI model or tool right? Like meta data as opposed to the actual physical model? Have you looked at future tools? It’s like an AI tools repository you can search within.
@ronskiuk Жыл бұрын
Thanks for posting. This is generating some fantastic results, combining it with the Keymate plugin it generates useful research and analysis
@StoutProper Жыл бұрын
What’s keymate?
@ronskiuk Жыл бұрын
@@StoutProper it's a plugin for gpt4 for internet access
@BirgittaGranstrom Жыл бұрын
I haven’t read the paper deeply enough but I wonder (and thing) what you are mean is that ToT is a way to describe how GPT-4 is thinking / reasoning “behind the scene,”?
@NeverCodeAlone Жыл бұрын
Very nice video and inspiration. Thx a lot.
@nekoeko500 Жыл бұрын
I need to test this method in smaller, local models.
@mahmoudaljuaidi Жыл бұрын
Love this video!
@micbab-vg2mu Жыл бұрын
Great it works. Thank you.
@TheSacredGrove Жыл бұрын
Nice my guy 👍🏾😊
@chrismachabee3128 Жыл бұрын
Well, still brilliant see. I saw something about the TOT, didn't dive in. Glad I didn't, I really liked your demonstration, and I can see and understand your conceptual presentation which is important to me, to know what you are talking about. Really good show. Thanks.
@elpato34 Жыл бұрын
Now this please Reflective Linguistic Programming (RLP): A Stepping Stone in Socially-Aware AGI
@alertbri Жыл бұрын
The paper titled "Reflective Linguistic Programming (RLP): A Stepping Stone in Socially-Aware AGI" proposes a method for building training data at scale for Reflective Linguistic Programming (RLP) which would be useful in developing socially-aware AGI. The proposed method is an extension of GPT that would be able to provide the AGI with the ability to reflect on its own thought processes and language usage, leading to more ethical and responsible decision-making. The paper emphasizes the importance of social awareness in AGI development and discusses the challenges of identifying early predictors of language abilities in humans. Have you been following David Shapiro's work on his Heuristic Imperatives? I love the simplicity of his approach.
@elpato34 Жыл бұрын
@@alertbri No, I am gonna go check it out right now
@dreamphoenix Жыл бұрын
Thank you.
@elmehdiezziar Жыл бұрын
Wow, thanks ✌️
@DefenderX Жыл бұрын
Maby this could determine the ultimate question of: Should the toilet paper be faced towards the wall or away from it?
@iamnickthegeek Жыл бұрын
Away from the wall. This is known. I have spoken.
@john_dee1431 Жыл бұрын
Interesting how when using this approach, GPT-4 can backpedal and advocate what it didn't originally. Let that be a lesson about strongly-engrained human biases.
@JcMETAV Жыл бұрын
so good!
@psychxx7146 Жыл бұрын
*Uses Xpapers plugin with the prompt template so that ChatGPT learns by itself
@AI-antics Жыл бұрын
Hi fellow prompt engineers I am building a full novel with GPT4 at the moment. I have been working on it for the past four months and I have made significant progress with it. I have the full writing style, the world-building, the plot summary, the character list, and the visual imagery for each scene. Obviously, I am massively exceeding the token limit and had to build it in sections. I am struggling to get the bot to produce any useful writing, but I can't think of any more information to teach it to produce good work. I'd love to hear your ideas!
@PazLeBon Жыл бұрын
i cant see anyone buying ai books, why would they? if they decide to ignore real authors they can get ai to write them a book from scratch, with thier own subject matter. they dont need you. or I anymore
@arielperez2445 Жыл бұрын
Hi, I am doing the same. Chatgpt can actually make some very interesting books, but it took me a while to get GPT 3.5 to continue the story as I wanted, at the beginning the token limit was one out of many drawbacks, but the more you practice, the more you get creative and find ways to overcome them. As to why someone would write books, the first one I did was a script for a videogame, but I liked it so much, that I continued it as a book. The second was a topic of my interest, this time with GPT4, so I just want it to read a fictional story about it, it's way different to the first one in which I was guiding all events, the second the AI is 99% the author.
@StoutProper Жыл бұрын
I’ve done a bit of work on this, pretty successfully imo. What is it exactly you’re trying to get it to do? What seed/primer prompts are you using? The general process I came up with was: Genre prompt > Author prompt > Story Template prompt > Story Outline prompt > Story Element Prompt > Character prompt > Plot detail prompt > Chapter Outline Prompt > Scene Outline Prompt > Scene Prompt > Dialogue Prompt > Review Prompt You don’t always necessarily have to use all the steps, but once I’d got the process worked out I found I could really quickly churn out pretty decent writing at an amazing rate. Obviously it needs editing etc, but there was some good stuff there. Here’s an example: A soft breeze drifted through the open windows, carrying with it the scent of freshly bloomed flowers from a nearby garden, mingling with the intoxicating musk of the city. It wrapped around them with a delicate caress, heightening their awareness of each other, and teasing their senses with a promise of something more. Sarah's fingertips grazed the edge of a document, her touch lingering just a fraction longer than necessary. Their hands brushed accidentally, sending a jolt of electricity through their bodies. Her fingers trembled as they grazed his skin, her touch both hesitant and hungry. James's breath hitched, his gaze fixated on her hand, the simple act stirring a longing within him that threatened to consume them both. The tension between them reached a breaking point, and they knew they had to find release. She leaned in closer, her voice dropping to a whisper. "But you know, James, judging by how much you’ve got an eye for my figure, I think we could make an even better team outside of work." She leaned back, her eyes locking onto his, her fingers tracing slow circles on the back of his hand. “What a pity we’ve got to keep it professional,” she added teasingly with a pout, drawing back slightly and giggling softly.
@EyeColorAnalyzerUnika Жыл бұрын
Chat gpt can rewrite and improve on your prompts. It's already replacing prompt engineers.
@maheshBasavaraju Жыл бұрын
Just wow! No need of any new models. Prompt approach itself is like an update to models in that it performs better and provide better solutions.
@jbexta Жыл бұрын
What would happen if you train a chat model on outputs generated from techniques like this? Would the model output better results? Then what happens if you do tree of thought on the new model?
@StoutProper Жыл бұрын
Yes, and it would give even better responses
@thecoffeejesus Жыл бұрын
I’ve been using this exact process without knowing it. Huh. Cool
@dausdk Жыл бұрын
We are teams
@felores Жыл бұрын
wrong paper link
@morris5648 Жыл бұрын
WOW?
@mcombatti Жыл бұрын
Love how I give a public lecture on AI "chain of thought" expanding comprehension with these individuals, and suddenly a paper "tree of thought" is released two days later 😜 my next research release coming this week, and I'm releasing the publication before public lecture this time. 😜
@BirgittaGranstrom Жыл бұрын
Its how it works when you are in synk with the universe😊
@StoutProper Жыл бұрын
Great minds and all that. I think tree of thought was originally conceived in the 50s by one of the godfathers of AI. We’re just going back to it’s roots. It wouldn’t surprise me if someone used GPT to come up with the idea to use it in all honesty.
@bmatichuk Жыл бұрын
I don't see how this backtracks. it just looks like chain of thought prompting with a selection. I think to make tree of thought work you have to write code.
@bztube8883 ай бұрын
It's great but it's called communication, not engineering. It doesn't mean it's more simple, but if you go to a library to study this, you have to go to a different building. Why do we pretend we reinvented the wheel just because we communicate with an algorithm, not a human?
@charlesmendeley9823 Жыл бұрын
Are you sure you reference the right paper?
@Candyapplebone Жыл бұрын
There are at least two papers on this
@maheshBasavaraju Жыл бұрын
I wish I could think as clearly as this tree prompted gpt😔
@soisas2811 Жыл бұрын
this looks like chain of thought, not tree of thought?
@EyeColorAnalyzerUnika Жыл бұрын
But ChatGPT can prompt engineer, right? You can ask it to improve your prompt.
@shemkipruto4394 Жыл бұрын
Can't you just ask chat GPT to tell you what tree of thought is, then after give it a problem to solve by following the steps involved in tree of thought process?
@pablobarquin Жыл бұрын
Sad you couldn't find a better example to test it
@TobiasRavnpettersen-ny4xv Жыл бұрын
Norsk eller.?
@noname-zg8lh Жыл бұрын
This isn't impressive. At all. How were these probabilities calculated? These success percentages? What math produced them? A person who has the ability to craft such prompts certainly has the ability to reason these responses. What magic is there in these responses? I see little.
@kodfkdleepd2876 Жыл бұрын
So it basically tells one everything they already know if they have any common sense... great. I still haven't gotten chatGPT to do anything beyond what I can already do. So far it is only a nice "interface" in to the large data sets of the internet(which happen to be mostly full of B.S.).
@avi7278 Жыл бұрын
This seems like your own interpretation of ToT. I don't see how this improved anything. In fact there was no objective thing to measure. Not a very good example.
@fotobanky-prodejfotekavide3859 Жыл бұрын
u can add info about your boss, who she/he is. Age, skills, type of person etc.....
@apophenic_ Жыл бұрын
This is just common sense being labeled for clicks.
@klammer75 Жыл бұрын
Think I’m going to put this into a LLM chain in langchain as a function and then pass this to an agent as a too, and see what happens….you should try the same Kris and let’s compare notes!🤔🦾🤷🏼♂️