🚩With all advancements of #ai around #llms they cannot be used in any decision-making scenario.

💥Remember hallucination is not a bug but the b product (feature) of #llms no matter whether you want it or not. This makes them 💯% unreliable for decision-making scenarios.

Sure, you can use #llms to extract intent or other use cases that were originally for #nlp, or to create #chatbots, and so on.

…but…

❌You should not use it for decision-making.
❌It is not a replacement for rules and heuristics models.
❌It cannot be used for prediction, forecast, classification, or abnormality detection.

So, what should you use to utilize #ai in decision-making?

💡This is where the Open Decision Intelligence Platform (DIP) comes into the picture.

The promise of the #DecisionIntelligence platform is to provide a set of capabilities that enable organizations to leverage multiple #AI techniques (including #llms) in a decision-making scenario.

How?

✨The Open Decision Intelligence Platforms provide below core capabilities
* Decision modeling and execution
* Orchestration modeling and execution
* Decision logging, audit, and composability

What do you need to look for when choosing a Open Decision Intelligence Platform in addition to the above generic capabilities?

Not all of the vendors in this space have everything for all your use cases requirements.

Here 👇 is the list of what you should look for in evaluation:
🔹 What type of decision modeling is supported? Proprietary vs. Open Standard e.g. (Decision Model and Notation – DMN CL3)
🔹 How can you extend the platform's capability?
🔹 How advanced is the orchestration design and execution? i.e. stateful and stateless.
🔹 How do you train #MachineLearning or import a trained model, e.g., PMML, ONNX, Python, etc.?
🔹 What type of data connectors and adapters are available?
🔹 What type of data formats are supported? JSON, XML, CSV etc.
🔹 What are Debugging, Simulation, and Testing capabilities?
🔹 What type of rules modeling is supported?
🔹 If you want to move away from the vendor, how can you still execute your decision, rules, and other models? (vendor lock-in)
🔹 How is the model's interoperability enabled?
🔹 Is there any on-the-fly data flow and processing capability?
🔹 Does it support both real-time and batch out-of-the-box?

These above questions put you on the right path to select the right vendor in #DIP category, but also there are also certain questions regarding security, roles access and etc. that you may want to know depending on your organizations need.

If you have any questions, comment below 👇

–
Follow me if you are looking for unfiltered views into #decisionintelligence #ai. Hit 🔔on my profile and get a notification for all my new daily posts.

Posted here.

Published October 22nd, 2024 at 07:30 am