Alle Videos

    KI-Agenten in einer Minute selbstverbessernd machen

    Ein Ein-Minuten-Muster, mit dem jeder Agent den eigenen Prompt kritisiert und verbessert. Demo in Langflow, auf andere Plattformen übertragbar.

    KI-Agenten in einer Minute selbstverbessernd machen

    KI-Agenten in einer Minute selbstverbessernd machen

    Wichtigste Erkenntnisse

    • Speichern Sie Nutzerpräferenzen in einer einfachen Textdatei, die der Agent lesen und schreiben kann.
    • Injizieren Sie die Datei in jedem Turn in den System-Prompt, damit Präferenzen bleiben.
    • Das Muster funktioniert auch außerhalb von Langflow auf anderen Agent-Plattformen.

    Verwandter Leitfaden

    Agentic AI in n8n Multi-Agent-Workflows
    Vollständiges TranskriptAnzeigen

    Aus Auto-Untertiteln bearbeitet und lesbar gemacht. Entspricht dem Gesprochenen im Video.

    Das Transkript stammt aus den YouTube-Untertiteln und wurde für die Lesbarkeit in Absätze gegliedert.

    In this video I'll show you the easiest way to create an AI agent that improves itself using your feedback and it is so simple it will blow your mind all you need is an agent a prompt and be able to write and read from a text file yes ladies and gentlemen a text

    File we're not going Fancy with our scalable kubernetes database cluster we just need a simple text file now here's how it works we prompt the agent to detect user preferences and write them to file then in our system prompt we will always read the preferences from the file so they are always up

    To date again we are basically delegating all the work to the agent so we're telling the agent hey if you detect some user preferences be it implicitly or explicitly stated by the user save them into this file and now then our workflow is set up in a way that it will always retrieve

    The information from the file into the system prompt and that's it so let's put it to the test let's ask a question why do humans need oxygen great now let's pull up the text file next to it so you can see the agent writing to the file directly now let's give it some

    Directives or preferences let's say always give a Hau style answer and bada Bim bada boom it recognize the preferences and roll it directly to file now let's say always finish your response with an emoji and again it recognized it and updated my preferences now to really make sure that it worked we will

    Create a new chat window without any any message history and now I won't say any preferences but it should load them from the file and adjust the response accordingly so fingers crossed and it works like a Chum anyway that's it for now you can apply this principle anywhere to quickly upgrade your agent

    To a self-improving agent which listens to your feedback and if you want to see more about agents or any educational cont content let me know in the comments below until then bye-bye and have a great day

    Auf YouTube ansehen: https://www.youtube.com/watch?v=i9s2oDi9HsE