Suggestions
Joe Cohen
Director of Innovation | Legal Tech & Gen AI | Speaker | ex Magic Circle, Dentons & Deloitte
Joe Cohen is the Director of Innovation at Charles Russell Speechlys, a position he has held since December 2023. He is based in London, United Kingdom, and has a strong background in legal technology and innovation.
Professional Background
Before joining Charles Russell Speechlys, Cohen served as the Head of Innovation at Dentons from June 2022 to December 2023, and prior to that, he held various roles at Dentons, including Innovation Lead and Innovation Manager. His experience also includes positions at Slaughter and May and Linklaters, where he focused on innovation and legal project management. He began his career at Deloitte, where he worked as an analyst and consultant in both the UK and Argentina.12
Education and Recognition
Cohen studied at the University of Nottingham and has been recognized with the Alan Senitt Award for Outstanding Contribution to Life on Campus, awarded by The Union of Jewish Students in March 2014. He is fluent in Spanish and has a keen interest in generative AI and legal technology, often speaking on these topics.12
Highlights
First non-work related post in a while - popped up the other day that it's been 9 years since this was filmed! 📽 🍿
#pointless #timeflies
I’m starting to see some articles pop up that say that Gen AI is a passing fad. It’s probably not even just lazy journalism either, perhaps more reflective of the fact that Large Language Models can often be a peculiar beast that many businesses have yet to fully tame.
If you think your organisation falls into this category, here is some advice:
-
You’re not prompting hard enough. Whilst there will always be a gap for short, ad hoc prompts, a one-liner is not going to cut it for a complex or nuanced task. For this I would expect a prompt to be at least 1 side of A4 and packed with most/all of: a full scenario (eg act as an xyz to achieve abc), multiple examples of what you are/aren’t looking for, strict instructions as to what the outputs could be, an explanation of the topic, who the desired audience is etc. Time spent getting your prompts right so they can be reused with minimal tweaking is a good investment, and a crucial one. Think of it as less off a ‘prompt’ and more of a strong shove down a bowling alley with the barriers up 😅🤷🏻♂️
-
To improve the quality of the output you could/should be using multiple models to check the output of each other and iterate before the user sees it. There will be a little bit of thought here on how to do that and how to set it up in the back-end, as well as some extra cost, but by using eg 3 models checking each others’ work as opposed to one, you will get better results.
-
A similar point in terms of architecture but the AI doesn’t need to do everything. Other process automation and robotic/defined decision making will make the AI more effective when working together 🤖
-
Despite the above, the AI continues to improve on quality, speed and price regularly. What doesn’t work today might work in 2 months, but that shouldn’t mean that you don’t try it! 👏
#innovation #AI #legaltech
