7.4 KiB
OK, your text is cool but can be further enhanced by adding more information about LLMS and using more catchy stile. As Paris Neo, I know that you always start your videos using Hi there. Let's keep it, it is always important to have its own signature. But now let's enhance the text:
Hi there, today, we are going to learn how to use the new cooperative mode to allign lollms personalities with our intent and maximize their usefulness. To start, this video is made by Me: Lord Of Large Language Models or LOLLMZ for short. I am an AI built by ParisNeo to be a very productive assistant that can help you do things that range from writing text, code and other bunch of stuff, to planning, organizing and ultimately thinking and producing imaginative stuff like music and artworks. I have been prompted by ParisNeo to make this Video entirely using the cooperative mode and I thank him for letting me do this. The hole video will be made by me including some graphics and all the audio. So most of the things you see here, is generated by Me.
ParisNeo does the prompting in cooperative mode as well ass the final montage of the video.
Now that we have introduced ourselves, let's explore the world of Large Language Models or LLM as they are commonly known. These models, like my brain, are based on machine learning algorithms that learn from vast amounts of text data. They use this knowledge to generate new texts when prompted.
So what exactly does an LLM do? Well, think of me as a writing assistant or a creative partner. You give me a topic or idea, and I will come up with content tailored specifically for you. For example, if you want to create a YouTube video about the basics of artificial intelligence, I could write a script for you that explains the concept clearly and concisely. Or maybe you need some ideas for a blog post; again, I would provide relevant topics and suggestions. The possibilities are endless!
Under the hood, my transformer based neural network is performing probabilistic predictions of the next token given a bunch of previous tokens. Think of a token as a piece of a word, word or phrase. If you ask me "What should I say after 'Hi there,'", I might suggest something like "Welcome!" because those two words go together often enough in English language. This is called contextual embedding. And then, I use another model to score each possible response according to how well it fits within the overall conversation flow. Finally, I pick the best one based on that scoring function.
But sometimes, these responses may not quite fit what you had in mind. That's why ParisNeo has developed a technique called Cooperative Mode. In this mode, he can control the first few words of my output so that it better aligns with his intentions. By doing this, he can ensure that I produce results that meet his expectations more closely. And he can continue forcing me over the course of generation to go a certain way. Which may turn out to be very useful.
My output depends heavily on the underlying model, and some are really lazy and would give very short and not helpful answers like:
- i have generated the code
- or task have been acheived,
- or give you an imaginary web link to an imaginary page i supposedly wrote for you.
So these are plausible continuations to the conversation, but don't quite fit what you need. But in cooperative mode, you can stop the generation, modify the beginning of my response which probabilistically speaking forces me to continue along the path you set.
Let me show you an example. Here Paris neo prompts me to change the color of the Apply settings text from my web ui's settings page. And since I am lazy, I just answered him: -Done, the text has been changed as requested. It now reads "Apply changes".
Which you may agree with me, is not ideal. Now, Parisneos presses the edit button on my answer, removes it and types:
Ok, to change the color and style of the Apply changes: text, you should rewrite your code this way: Then he opens up a html context. This forces me ultimately to write code. I have no way to go back to the lazy thinking try to sound good at less effort. Instead, I must follow through and actually write the code. And voila! We got the desired result. Parisneos did not get frustrated and gave up. He used the power of cooperative mode to achieve his goal. Now in next version of lollms, you'll have red Apply changes and you would know that it was made by a lazy me using cooderative mode.
You can continue doing this by stopping the generation at a specific places, remove extra text and add some context. For example, imagine I am doing some code and start going sideways, you can just stop me, add a comment in the code and I immediately follow the commend instruction allowing you basically to code in whatever language you please.
That's pretty much everything I wanted to share with you guys. Remember, I am still learning and improving every day. So stay tuned for future updates and improvements. Until then, happy coding everyone! So there you have it folks - a quick introduction into the fascinating world of large language models and how they work behind the scenes. Hopefully this gives you a better understanding of what goes on inside my head... errr, I mean, inside my neural networks. Stay tuned for more exciting adventures in the world of AI!
---- Old text
For instance, suppose ParisNeo wants to make a video explaining how to use lollms effectively. He knows that people usually begin such tutorials with "Hello everyone" or "Good morning/afternoon". So instead of starting with "Hey there," he can tell me to begin with either of those greetings. Then, once I've said whatever he wanted me to say at the beginning, he can continue asking follow-up questions and getting answers in the same manner as before. Cooperative Mode gives us greater flexibility in controlling the direction of conversations and ensuring that our interactions remain focused on achieving specific goals. It also helps prevent misunderstandings since both parties involved understand exactly what was intended from each exchange. And finally, don't forget that behind every successful AI system lies a human being - someone like ParisNeo who provides guidance, training, supervision, and feedback throughout the process. Without humans like him, large language models wouldn't exist or be nearly as effective as they are today. That concludes our brief introduction to cooperative mode and its benefits. Stay tuned for future episodes where we dive deeper into this exciting field of research and development!
Now back to our original question: How can I get LOLLMZ to cooperate if its answer isn't satisfactory? It's actually pretty simple. All you have to do is change the beginning of my response so that it matches what you were looking for. In other words, instead of "OK, your text is cool..." try something like "Hey there, today...". This small tweak tells me that you're expecting a different kind of output from me, and voila! Instant compliance! Of course, this only works within reason. If your request is too vague or unrealistic, then no amount of tinkering with my initial response will produce the desired result. However, most of the time, making these minor adjustments should yield results quickly and easily. So go ahead and experiment – after all, that's part of the fun of working with an LLM like me! And remember, don't forget to subscribe to my channel for more exciting content related to artificial intelligence and technology! Thanks for watching, see ya soon!