NOESIS 用 Jev 做资产筛选与持仓监控

NOESIS @noesis_network原推发布于 2026.09.21

在 X 上打开原推

Some notes from our latest experiments. JEV has changed a lot about how we’re building NOESIS. LLMs are good at deep investment research over longer time horizons. They can use a harness to gather information, investigate a thesis and work through a complex decision. The process usually starts with a goal, then a plan, research and a conclusion. JEV fills a different role. It doesn’t chat or write code. It makes judgements, quickly and cheaply. That changes the game. Large volumes of information that were too slow and expensive to process entirely with LLMs become practical to evaluate. We’ve been testing JEV in asset selection: scanning unstructured information and combining those judgements with statistical data to identify tokens worth investigating. In our tests, JEV surfaced promising meme candidates at an early stage. We hadn’t expected it to be this useful there. Another experiment is position monitoring. A traditional risk system can check positions every minute, but its decisions usually depend on fixed thresholds. In our experimental logs, JEV has repeatedly challenged stop-loss and holding-period exit signals, judging the current drawdown to be temporary. That judgement still needs to be tested against what happens next. But it offers a way to bring context into frequent risk checks. Asking a conventional LLM to review every check could cost more in inference than some of our small positions are worth. These are practical experiments we’re bringing into NOESIS, not a finished system. We’re still learning how the pieces should work together. NOESIS isn’t a backtesting demo operating in an ideal world where even gas costs are ignored. It has to enter the battlefield from day one. We may be one of the first projects publicly experimenting with its role in AI investing, and the results so far have given us much more to explore. JEV also opens up another possible business model for robinhood:0xd25efceb9d690e2a75a8b75e52a7c63cb9a214e0. The info filtering and risk assessment systems we’re building could become services that other agents and applications can use. Real investment decisions inside NOESIS would put those capabilities to the test, with outcomes feeding back into how they improve. That gives us a way to build a track record for the services themselves.