Start Your Design Session
Your Socratic Uptake Assistant is ready: Enter your hypothesis, describe your model system (or paste text from a paper). Press the button only once. After the interview is finished („Do you already have an idea or concerns?“), it will take a minute. The experiment engine is currently running on GTP5.1 and its MASSIVE, so patience! Run your hypothesis through a few times, with only small changes, you will get different alternatives! I can discuss them with you. Get in touch.
Reasoning As Navigation (RAN)
AI can be trained to follow a reasoning path, if you provide them with an explanation of how to do it. For AI these explanations come in the form of operators – algorithms that direct resoning in loops with different constraints. This reasoning can be compared to how you navigate a sea if you have a chart, and light beakre to navigate towards. In Swedish these navigational routes are called „farled“. When I use linked operators to guide AI from input to experimental design I call that Farled. This farled will design experiments. Other farleder do other things.
Please write a comment on how this worked for you and leave an email if you want updates (I may use your input for refinements and then I can share those)
Do you have questions?
Contact
Reasoning as Navigation lithe: The experimantal designer
Socratic uptake assistant (SUA)
SUA will not only interview you but also analyse your paper and your hypothesis. In this SUA we are doing a very minimal analysis and currently you cant edit your input, so if you are unhappy, you need to start over.
Universal Causal Operator (UCO)
UCO is a logic layer between SUA and RAN which strictly enforces a policy to only change one parameter at at time, to introduce 3 tiers of controls, identifies the strongest competing hypothesis and designs a control for this hypothesis and quality controls the SUA output.
Reasoning as Navigation (RAN)
Here you see a small and lighweight version of the recursive reasoning engine RAN. this is the secret sauce. RAN will give you both a structured full JSON output and a description of the experiment. RAN will also suggest a next experiment for a new hypothesis. This new hypothesis can be cycled back into SUA if you are interested.
A very first experimental design engine
There is currently noone else on the market offering this type of experimental design, outside closed loop (automated) experimental systems. RAN lithe has a bigger brother that can be used to run RAN at scale. UCO can be modified to test for 2 changed parameters at the same time, not just for one as now. Here I use Gemini 2.5-pro as an engine, but ChatGPT5.1 works really well as well.
Schreibe einen Kommentar