KeyBot v2

KeyBot is a secure version of AI Chatbot that will not share any data that you enter with the outside world. Therefore, your prompts (what you put in to KeyBot) may include confidential information. You can access Keybot via Keyed-In under Support > IT menu.

KeyBot is an amazingly powerful tool and should form part of the way you practise law. But there are important things to remember:

1. KeyBot will get things wrong. Sometimes slightly, sometimes absolutely and completely. When it makes mistakes, it will not indicate any form of uncertainty in the way a human might, and will present utterly false things as if they are 100% correct. You must take care not to be lulled into a false sense of security by the confident tone adopted by AI tools. Sending out something drafted by KeyBot that is imperfect is not KeyBot's fault, it's yours; and we'll likely be sued for negligence should the client suffer loss. There might also be regulatory consequences.
2. You need to spend time practising with KeyBot before you use it for any client work. It is important that you work out what it can and can't do. You may be frustrated at a number of basic failings it will make. But you need to go through this process so you are under no illusions as to the variability of KeyBot's output.
3. Used properly, KeyBot can save you time and help you do more with less. In some cases it will help you do things you could not do before at all.
4. You should keep records on your file of everything. KeyBot is no exception. Cut and paste into an email to yourself (and then file it in NetDocs) any relevant working.
5. KeyBot is another work tool. The firm keeps logs of what you do and your use of it is not different from your use of any other firm tool. Consider how you use it, use it for work purposes and do so ethically.
6. KeyBot can access various large language models. Each model has been trained on data up to a certain point in time, and that "knowledge cutoff" varies by model version. KeyBot does not automatically know about events or developments that occurred after a model's training cutoff. You can see these dates when selecting the model. As newer models become available, we will make them accessible in the tool, but the specific cutoff date will always depend on the model you select. If your question requires information beyond a model's training cutoff, KeyBot will not be able to provide a reliable answer unless you upload newer resources.

Please also take the time to read the firm's Generative AI Usage Policy and be mindful of it when using KeyBot.

Where to start?

You should start with what you already know inside out. Start with something you did recently, something to which you already have a model answer. Ideally start with a simple task or just use part of some earlier work. KeyBot is better at smaller discrete tasks, especially tasks that relate to a single input, e.g. a document and doesn't require any thinking outside of that.

There is no instruction manual for ChatGPT. It can do just about anything, albeit with varying degrees of success. The best thing to do is to play around with it safely and solely for the purpose of working it out. There are thousands of YouTube tutorials and articles on it. By all means have a look at some of these, but accept that you can never know everything.

Use cases

What you can do with ChatGPT is limitless. Here are some examples:

• Putting in a PLC clause and asking it to change it for you in certain ways
• Putting in something you have drafted and asking it to critique it from the point of view of the other side or generally
• Putting in something before you start work on it and asking it for the key issues for you to think about
• Generating emails where you give it the bones of what you want to say and it turns it into a much longer well-structured draft
• Looking for inconsistencies within one or more documents
• Summarising long documents
• Proofing your drafts for you

Tips for use

• Be careful and keep all KeyBot's limitations in mind from the start. It makes mistakes and it can't weight trustworthy sources higher than untrustworthy ones
• Start a new chat (button in the top left) for each new thought. KeyBot will assume anything in the same chat is relevant. Starting a new chat wipes the slate clean and prevents confusion.
• Tell KeyBot what you want it to be, e.g. an employment solicitor in England and Wales. Remember it does not know you are asking it to be a lawyer unless you tell it. KeyBot is not a special lawyer's version of ChatGPT.
• Tell KeyBot very clearly what kind of an answer you are looking for. Give a framework and say what's important.
• Tell KeyBot the background to the task. E.g. sector, the aim of the client, who you act for (buyer/seller, employer/employee, claimant/defendant etc).
• Keep your instructions simple and logical.
• Split up longer tasks.
• Ask follow up questions when it does not produce what you were expecting.
• Ask the same question more than once and in different ways.
• Ask KeyBot for its reasoning and sources.
• If asking KeyBot to change things, ask it to do so in CAPs for new bits and to show at the end what it took out. Otherwise cut and paste into word and do a Track Changes Comparison so you can see what has changed.
• You can tell KeyBot what tone to use and to use English spellings.

How do I bill for things KeyBot does?

Once you have tested out KeyBot, feel confident using it and have a sensible task, offer a fixed quote for that task and use KeyBot as part (not all) of the work product. It should cost the client either the same as it would have cost had you not used KeyBot or a little less. If KeyBot saves you time (and it will) that should not be a saving for the client alone (or sometimes not at all), but you'll need to use fixed fees (on a discreet task is fine) to get there. You can't say things took you longer than they did. That's dishonest.