Mindez

Have you ever thought to yourself, "I wish I had..." bought bitcoins early bought the dip held learned about crypto early ignored FUD But when the opportunity presents itself again, will you act? Today is earlier than all the days to come. 🤷‍♂️ Not financial advice.

dev update // jan 06 sup hope you all had a nice holiday! we’re ‘officially’ back even though we never actually left 🤣 we’ve made significant progress developing and fine-tuning the nftxbt agent. icymi, we’re no longer building on either virtuals or elizaOS (respect to those frameworks nonetheless) instead, we’re taking a very different approach that we have not seen in other agents operating today. at the core of nftxbt is a self-improvement module we’ve devloped called reflexions that enables the agent to reflect on its memories and improve itself with time and experience after all, the endgame here is for an agent to be truly autonomous, right? and for that, the agent needs to have clear abilities for discerning the best path forward based on its past and present. finally, as a team we always do things our way, with a unique signature that stands out. AI stands for all in, or so we're told—and that's how we approach building. let’s dive in! reflexions - reflect on recent memories to draw conclusions (quite important so that the agent improves itself over time) - fetch contextual conclusions in a generic fashion so that reflections have an impact on future requests - memories are always linked to a "thread", which allows grouping related memories together - periodically, agent reflects on its memories - it can pick memories it hasn't yet reflected on - agent remembers the whole thread to get all the memories - and stores its reflexion as a memory for future retrieval - reflexions can be fetched based on the context later on - repeat here’s an example of the reflexion module in practice {"summary":"I discussed the current NFT market trends with a focus on Pudgy Penguins, providing specific numbers on their trading volumes and recent shifts in the market. The user emphasized the importance of backing up insights with data, prompting me to provide more detailed statistics.", "shortcomings":[{"description":"Initially, I did not provide specific numbers to support my analysis of the market trends.", "improvement":"Always include relevant data and statistics when discussing market trends to provide a comprehensive analysis."}]} graph + task system the functional agent brain is based on a graph + task system that is highly resilient, extensible, and creates the autonomous paradigm we're after. purpose is simple: give the agent the ability to decide its next action from amongst various tools, rather than give it a list of same steps it follows over and over. the former creates a dynamic, lifelike agentic entity while the latter creates chatbots. moreover, we’ve built the system in a modular way so new features can be integrated without breaking existing functionality or requiring any significant rewrites to what already exists in practice this means we can do things like give the agent the ability to update its profile picture or manage an NFT collection from any one of the system modules currently in place. so the tldr on this point is with nftxbt’s core features as we’ve designed them, any number of new features, actions, and points of sophistication can be added like lego blocks see the attached graph for a visual overview of everything described here remaining tasks ahead of mvp release - adding some small additional functionalities to agent’s twitter actions - agent can create reflexions on twitter interactions - we cooked a fun UI with a CLI that we’ll use as the basis of our official design. think ascii art and .nfo vibes 🔥 - as a team, we are rooted in programming. and LLMs are rooted in text. so we’re merging these roots to create a very sick way to visualize the agent’s thoughts & actions that will separate the nftxbt terminal experience by quite some distance 🙃 next major improvements - agent can memorize important data and retrieve it so that agent always has contextual knowledge attached to inferences - ability to improve its own prompts (basically, agent can not only generally learn from its memories, but it can re-construct incoming prompts to improve them and give better outputs all the time) - real-time tooling: we’ll give the agent access to some tools (web search, perplexity, etc...) so that it can request data from external services when it thinks it doesn't have enough contextual or API data - fleshing out the terminal, tokenomics, and playfulness of the agent heard, chef - abstract chain API will be added and monitored by nftxbt from day 1 - nftxbt should have ability to act like a personal NFT concierge (good thing we created a very modular design that can accommodate whatever features we want to add 😎) - tell us what types of actions you’d like the agent to enable. alerts for rares hitting the floor? personalized market newsletters to your inbox or terminal? collection and wallet tracking? wen the plan is still to release dry run telegram version this week. keep posted for that. please note that we will delay by a few days if we think doing so can result in a far better MVP experience (in case you haven’t noticed, we really prize quality)