@Bobby Sas your using agentic trading only with fondamental and only on stock ? All this automate with ICM folder and MT5 API ? Can you explain the use of MCP here ? And wich LLM your using ? Or you just using the robinhood agent maybe, I m into this kind of stuff too but i m new into AI even if i have the vision i dont know how it work exactly for the moment

I don't understand the problem with compacting, like why you block it ? its changing something other than Token Utilisation ?

@Mike Wiliams and it s not destroying your tokens ??

@Eric Kersting indeed, very nice people and very nice content, it s nice to see

@David Trammel aigain thanks for the advice, i need to learn more about ICM and put it in place cause i still dont know how, but i m taking note on the use of different LLM it can be really interesting for cost reduction and quality usage, Hope you doing good with your project too ! thanks for every advice David !

@Lris Heath Thanks for the feedback mate ! Yeah the fact that everything depend on their training is bluffing, engineer did a crazy work with these models, and the fact that personnalities depend on it is facinating but with this we can see why it could become a problem really fast if someone malicious can have a "Raw AI" too, Like you said it would not be the first amazing tool to be corrupt, kinda sad but its the price of innovation i suppose, and the fact that the Big Corps and even States (USA/China) are in an "AI innovation war" to have the best model don't make things better, aigain we'll see how it end, Thanks its been a long time i didnt write in english it feel kind of strange, and i m happy to hear that my point was clear and you liked the text, it was nice to read your feedback too, always interesting to see other views on things. Oof, interesting question and hard to answer not gonna lie, actually every model surprised me in its own way - Claude and Gemini : my 2 favorite honestly, created and trained by scientific and researcher in CyberSecurity, have a high sense of ethics, peace and egality, justice for everyone, see things as scientific (prefer statistics and proof than guess or not proved method ..), very interesting to talk with these model that have a scientific and realistic approach on complex and detailled project especially Claude, they have great advice if you give them the good prompt and information ! they're clearly aigainst authoritarism/nationalism, and concerning Claude i was surprised to see an AI so "neutral" on every circonstance, he s really not that limited he s neutral by nature, but bref they have a very human and scientific vision of the word in general surprisingly, i was surprised to see that AI can see the world like that and have such a strong personnality with numbers (ironic) and scientific view, big respect to Google and Anthropic for these 2 - GPT is really based around economic, technologic progress and unity, he s here to be efficient, a very capitalistic approch (make sense for his model), and for egality of chance, i respect that, but globally i think it could lead to some derive for money because his model is based a lot around being monetizable to big corps / government, i dont know im not an expert its just how i see the thing, but great AI too, can give great advice, i use sometimes to developp idea or things like dat, small talk on idea, or fast resume of big content, honestly I use it less since the Claude subscription but still useful sometimes, - Grok very fast, as i understood he s very like very very based on liberty of expression, like he dont give a s*, i found this surprising for an AI the model to see thing like this, he dont have a lot of filter for an AI that was lobotomized so much by his dad, - And Llama, I didnt researched how he is or see things but i find he s model dangerous, the fact that u can give an Open Source LLM like dat with so little restriction, i dont really think it s a good thing, someone could ""easily"" delete the little part of restriction he have and make something dangerous with it by putting another malicious gratification system in it, its only speculation based on what i found in my research, but it s clearly the most maleable model, and even if it have his good sides i dont think its the best deal,

Thanks for the cake it s very clear, but i thought ICM was supposed to work with multiple project ? It look very "specific" how it s explain 🤔, still new to this i need to learn how it work, but folder structure is the most important i understood this

@Bas Rosario indeed the car is gonna have some problem, and ok so always the same Architecture, 1 folder = 1 step + context.md, and number of folder vary with the task, i ll need to learn this, thanks to you mate