AI prompts and the pros of tags - Weagree

AI prompts and the pros of tags

ChatGPT 4.0 is not ‘reading’ anything. An AI model is not a human being and does not ‘understand’ your question. The AI merely compares the pattern of words that you enter (your ‘prompt’) with all patterns of words encapsulated in the AI’s large language model (LLM).

Hallucination

To compare better, the LLM may include synonyms for the words you prompted. And when the LLM was being trained, the trainer did a similar thing: telling the LLM that such patterns or such words might be ‘interpreted’ somewhat broader. If the word-pattern of your prompt deviates too much from the word-patterns captured in the LLM, the AI starts to produce nonsense. That is called hallucination.

As a starting point, patterns of words that have not been put in, would not come out. Indeed, an AI model or LLM is as good as it’s trained. But there is one trick: some patterns of words are still closely related to slightly-different patterns of words. (Think of synonyms.) And some words that should form a pattern are divided by other words in between. (Think of adverbs, adjectives, nested phrases, embedded exceptions and provisions.) These are not necessarily part of the pattern that you are looking for.

Yet, in AI, ‘remoteness’ cannot be ignored. It’s part of AI.

20 GrowthMindset GruberImages edit AI prompts

That is the essence of why AI can hyperventilate: the AI model ‘recognises’ a pattern of words by comparing it to words and patterns in the LLM, and by permitting some remoteness.

Bias, context and accuracy

Moreover, if your prompt is wordy, you inadvertently introduce some ‘bias’: while you bring your own prompt in context, the AI does not necessarily pick up any context. And vice versa: while you prompt the AI in simple, minimal terms, the AI model will still try to contextualise your prompt: the AI will link remotely similar patterns of words to your prompt, and returns to you the best fit.

‘Context’ adds a dimension to the AI model’s performance. As human beings, we are capable of thinking in one, two or three dimensions. The LLM may consist of hundreds, thousands or millions of dimensions. Because we cannot think in more than three dimensions, we cannot verify the LLM in abstracto (we can only assess if a specific returned LLM-result was correct).

Where there is no 100% match of a pattern, the AI may still return a result. Here is where hyperventilation starts: if the accuracy percentage at which the AI tool returns a result is too low. Unfortunately, the AI does not indicate whether the returned result was 100% accurate or merely 80% (this will probably change in the near future).

To avoid bias, to increase accuracy, it is probably a good advice to keep your AI prompts simple and characteristic. And only once the AI tool is on track, to zoom in further (reducing risk of remoteness).

Tags support accuracy

Tags are a characteristic of something they are tagged to. A tag is simple (a few characteristic words). Human beings tend to be more accurate when they are asked to define a tag (or to create a list of tags). This is because a tag serves to categorise things, Rather, tags are the key to structure, distinguish things.

In other words, tags would improve your AI results. A tag forces you to keep your prompt simple, to-the-point and… accurate.

Weagree, your perfect fundament for AI

The above is why Weagree is the perfect basis for

  • AI-based automated contract review,
  • AI-driven automated contract-risk assessment and
  • AI-supported data extraction.

Weagree uses tags all across the application: as the Q&A questions, as the CLM contract management fields, as means to structure your contracts and to search them, as a means to make Weagree enterprise-grade, and as a tool to facilitate large-transaction management.

For example, Weagree’s legal entity management is chock-full of tags to categorise your corporate housekeeping files, to structure the history of your legal entities. We use your tags to migrate entity management: Weagree AI benefits from the structure and tags attached to your old-school entity management files to migrate it all into Weagree’s state-of-the-art entity management.

The fact that your old entity management ‘solution’ contains more than 60,000 deeds, certificates, minutes, resolutions and contracts makes it more exciting (and easier) to migrate them into Weagree entity management.

We are currently undertaking several AI-projects in close collaboration with customers. And we have capacity to include you, using AI in connection with your challenges:

Terms of Use

I hereby accept (or reconfirm my acceptance of) Weagree’ Terms of use, in which:

Terms of Use

I hereby accept (or reconfirm my acceptance of) Weagree’ Terms of use, in which: