Hi
Happy Monday! Hope the weekend was good and that the week ahead is shaping up well. A few notes from the world of AI to go with your morning ☕:
EU Agreement Advances AI Rules Amid Deepfakes Urgency
Europe’s new rules for AI, proposed in December 2023, have taken a large step forward this week with EU countries agreeing on the technical details of the Act.
France (home to Mistral AI) had been the primary blocker with President Macron warning in the past that such “regulation must be controlled, not punitive, to support innovation” and new laws should “regulate the uses, rather than the technologies themselves”. It appears that the deepfakes of Taylor Swift and Joe Biden (outlined in this post last week) have underscored the need for these new rules. Next steps: the Act goes to a committee of lawmakers mid-Feb and the European Parliament to vote in either March or April with the act coming into force before the northern hemisphere summer.
In an interesting counterpoint, the UK’s House of Lords has taken a potentially more “open” approach to AI regulation, advocating for parliament to “focus more on opportunities” while tackling near-term risks and carefully investigating more catastrophic risks. Will be watching with interest as this situation develops.
Opting Out of AI Learning from You: ChatGPT Gives You a Choice
I’m all for AI – at any given moment in the day, I might have 3 open working on as many different tasks. But the fact that these systems are hoovering up all that information, using it to improve themselves by learning from my real personal data – when I think about it, it gives me the creeps.
Leon Furze (AI thought leader, author, and one of the key developers on the excellent AI Assessment Scale) is absolutely on point here: “The default for privacy should always be opt-in. It never is”. The question is, as AI systems like ChatGPT rapidly advance, who decides what personal data they should use for self-improvement?
New OpenAI opt-out settings finally give you control. You can now prevent your data from being used to improve the AI model going forward. No more indirect tutorials or vague explanations. Go here and follow onscreen instructions – OpenAI Privacy Request Portal.
Don’t Miss: AI Innovation Symposium for HE Educators (Weds 7th Feb 9am AEDT)
This Wednesday 7th, don’t miss the 2024 AI in HE Symposium – Australia and New Zealand. The University of Sydney will host talks on creative applications of ChatGPT, Copilot, Midjourney and more to enhance pedagogy. Presentations and breakouts will spotlight practical integration examples like:
- Using AI for personalised feedback at scale
- Generating images to boost text analysis
- Building VR therapy practice scenarios
- Curating AI co-writing experiences
Sessions will reportedly balance inspiration and adaption advice – we’ve been fortunate enough to have one of the key organisers in Danny Liu speak to our staff at RMITV so, on the basis of that alone, it should be amazing. If you’re eager to explore thoughtful ways to employ generative AI as a teaching and learning multiplier, this free half-day event is an excellent community builder and resource.
What happens to grades when students use AI?
Navigating the Jagged Technological Frontier: Field Experimental Evidence on the Effects of AI on Knowledge Worker Productivity and Quality (Dellacqua, 2023) is a fantastic paper – demonstrating enormous productivity and quality gains for “typical” knowledge worker-style tasks. I’ve been looking forward to seeing more refined, subject-specific studies for other areas. With Choi, Monahan, and Schwarcz (2024), lawyers appear to be the first out the gate and it seems that the results largely align with Dell’Acqua et al.
Choi et al found a levelling effect where the output of weaker-performing students using a GPT4-based AI was improved to near that of the higher-performing students (see below). Otherwise all participants, regardless of level, enjoyed significantly increased speed and satisfaction from using AI to complete legal tasks. Looking forward to seeing more studies along these lines as the rollout of built-in platforms like MS365 Copilot and Google Bard improve and the year progresses… 👀 🍿
Quality gains – Choi et al (2024)
Efficiency/speed gains – Choi et al (2024)
Two giants are slowly starting to move… Google and Apple
Remember a few weeks back when we heard that Gemini, Google’s new model was reportedly ‘marginally’ better than GPT4? Underwhelming at the time but the fact of the matter is Google has some of the deepest reserves of data in the world (e.g., YouTube) so it would probably ramp up quickly once it did. That seems to be happening – Google Maps is getting “supercharged” with AI. Details are still low but, if you combine location history with real-time occupancy and tell it “find me a good [cafe/bar/restaurant/local attraction] that’s nearby and not too busy”, it could probably give you pretty solid, personalised, and real-time advice. Madness
And, fresh off the launch of the Vision Pro last week (which looks awesome btw), OpenAI are jumping on the opportunity to get into bed with Apple – launching the native Apple Vision Pro ChatGPT app. In the quarterly earnings call, Tim Cook also noted that Apple is putting “tremendous time and effort” into integrating AI with its platforms – aiming to release “later this year”. That could be really interesting – or at least hopefully Siri gets a lot more useful. 😉
We hope this edition of the newsletter has been of interest to you. If you’re new here and it’s been useful, please do click subscribe and you can expect a weekly update every Monday from now on. If you’re already a subscriber – thanks for your ongoing interest and support! Either way, if you know others who might benefit from reading this weekly, please forward it on to them too.
Otherwise, and in the meantime, to everyone celebrating the Lunar New Year, chúc mừng năm mới, 春节快乐, and happy lunar new year! All our best wishes to you and yours for the Year of the Dragon! 🐉
All the very best for the Year of the Wood Dragon