This document contains mostly very general provisions. It is intended as a source of inspiration for those drafting laws and other forms of regulation in any regulatory or policy area. Readers are invited to select provisions that they consider useful for their specific regulatory or legislative task and to adapt them to their specific needs. No provision should be taken without considering the need for adaptation.
The document is not intended to be a universal generic model law. However, in the absence of a sector-specific model law, it can be used as a basis for drafting legislation and other forms of regulation. If you intend to use the document in this way, you should be aware that it contains redundancies, partly because it is derive from very different legal traditions. For example, in some jurisdictions, interpretive rules are placed at the beginning, while in others, at the end. To reflect this, we have placed scope-related interpretative provisions at the beginning and at the end whilst recognising the interest in placing all interpretative rules in one place. Moreover, we kept redundancies as hardly any user will use the document in its entirety. Accordingly, certain topics appear at various places.
The document has been produced by two artificial intelligence programs known as Large Language Models (LLMs): Claude (Anthropic) and Perplexity. It is the result of a long series of instructions and tasks that took the human in charge several days to complete.the document was only marginally revised by humans. The order of the provisions, and the provisions themselves, could be improved depending on the jurisdiction. However, as its main purpose is to serve as a kind of quarry, the document can already be used to make draft legislation more complete.
The basic pattern of the document has been developed using the Regulatory Institute’s model laws, because these model laws integrate regulatory knowledge from many jurisdictions and sectors alike. They, therefore, provide an excellent international and cross-sectoral basis.
However, the LLM interface Perplexity[1] has, in many cases, suggested additional very sensible rules based on regulations that have been sifted through around the world. In doing so, Perplexity continued the Regulatory Institute’s approach of learning from regulators in all sectors around the world and making the knowledge gathered available globally and across sectors. It is this approach that has been continued by regulatory practitioners supporting the Regulatory Institute who manually redacted the text to a modest extent. The document is therefore the result of three rounds of international gathering of regulatory knowledge. Feel free to suggest further improvements.
This document is complemented by two lists that have been developed in the same work process:
- The list of powers/obligations;
- The list of sanctions and accompanying measures.
Inadvertently, the document proves that at least some LLM can be used to complement draft regulations and to make them more effective. If you want to embark on the adventure of LLM-based regulatory completeness, check or similar forms of revision, here are a few tips:
– Do not underestimate the time involved. Plan your work to avoid long interruptions, because controlling LLMs for large tasks is also complex and requires a lot of attention, oversight and memory on the part of the LLM user. For example, the ‘raw’ lists of provisions used in the three documents produced by this working process took more than 20 hours to produce without any review, and the light review took even longer.
– Test various LLMs. The LLMs we found to be particularly performing might not be the most performing tomorrow. If you do not have time to test various LLMs, seek advice from us. The most commonly known LLM (currently ChatGPT) might not be the best performing one.
– Communicate with LLMs as you would with a person.
– Sometimes, ask if things are clear or to rephrase the task.
– Be aware that the LLM, like a person, may want to be lazy and cut corners. It may also cheat a little here and there. Give clear feedback on this.
– If the LLM does not ‘know’ where to look for good provisions, refer to www.howtoregulate.org and the laws of New Zealand, Singapore, Hong Kong, Canada and South Korea. South American laws also contain interesting provisions. US and European Union laws have been digested by LLMs anyway.
– Make your instructions as detailed as possible. You can also ask the LLM to fine-tune the instructions in a first round before asking them to complete the task.
– Break up large documents into manageable chunks.
– Be careful about cumulating different tasks in one instruction – this often leads to unwanted results.
– Evaluate each interim result and give feedback. It is tedious, but worth it in the end.
– Ask for a redo if you are not happy with the result.
– Be clear about your objectives and the overall goal, because sometimes the LLM will identify useful intermediate steps that you have not seen yourself.
– Give another scheme as an example of format or structure.
– Check whether it is useful to refer to a model law developed or collected by the Regulatory Institute to outline the format or structure.
– For the content, indicate reference documents (e.g. existing laws or model laws) that should be used as basis.
– If your LLM cannot digest information from the internet (like the otherwise currently excellent LLM “Claude Sonnet 3.5”, which can also be accessed via a professional Perplexity subscription), upload the reference documents as pdf or txt files. For example, we have uploaded a slightly truncated .txt file with all the model laws from the Regulatory Institute. We can provide this file on request.
– Use the following document as a starting point or reference when instructing LLMs. Select the parts that you are interested in.
[1] Strictly speaking, Perplexity is not an independent LLM, but an interface that builds on various LLMs.