Hi
Hope you all had wonderful weekends. I floated back into the office on Friday morning after three days on motorbikes out on the border of 🇻🇳 and 🇨🇳 – still buzzing. If you ever get a chance to ride Ha Giang, I cannot recommend it enough – already planning the next trip 🤩🏍️. In the meantime though, a few headlines from the world of AI to kickstart your Monday:
The Jailbreaking Dilemma: Balancing AI Performance and Safety
In a week where concerning, almost dystopian reports about AI as a tool of war are coming to light, this is perhaps a good moment to touch on jailbreaking AIs (i.e., overriding the system guardrails that stop them from giving out private information, undertaking harmful actions, etc.). It turns out that the increased power and, in particular, context window size of AIs like GPT-4, Claude 3, and Gemini Ultra makes new jailbreaking techniques possible.
Anthropic (the team behind Claude) recently released a fascinating research paper which outlines a simple but effective way to bypass the safety features of LLMs that are designed to prevent them from generating harmful or illegal content – overwhelm it with hundreds of negative examples. Research paper here and more user-friendly overview here.
There are patches (e.g, adding a mandatory warning) that reduce the chances of jailbreaks but unfortunately that impacts the systems overall performance. This is an ongoing problem with no easy solution in sight as AIs continue to evolve – will be watching with definite interest.
“It’s brutal if you think, like, as a human”
Or maybe people are the problem. Case in point: you might recently have heard of Suumit Shah – the CEO of Dukaan, who recently bragged on Twitter about how much more efficient his business is now that he’s replaced 90% of his customer support team with chatbots.
A follow-up interview with Shah contained this spectacularly revealing exchange:
Interviewer: “Why did you do this? It seems a little brutal”
Shah: “It’s not, I think… it’s brutal – if you think, like, as a human”.
Cue Jon Stewart. In an fantastic and extremely timely piece titled “On the False Promises of AI”, Jon destroys the hype and shine around AI’s long-term promise (they will fix climate change, cure disease, etc.), dragging much-needed attention back to the disruption that is coming soon – or, if you were unlucky enough to work for Dukaan, now. Ahh Jon – it’s so good to have you back. Your bullshitometer is absolutely world class (tho what’s with blocking sharing on LinkedIn?).
Topically, US Senator Bernie Sanders is now calling for a 32-hour work week. Sound radical? It turns out that the 40hour work was established in 1940 – so 84 years ago… he makes a great point about the “modest” changes to the economy over the last 84 years… 😂 #hesnotwrong.
AI in HE – have we reached the Tipping Point?
Closer to home for all of us, recent developments in the United States and Hong Kong suggest that Artificial Intelligence (AI) is on the cusp of transforming HE. At Boston University, a dean proposed using AI tools to “offset” the impact of a graduate student instructors’ strike, sparking a heated debate about the role of AI in teaching. Meanwhile, the Hong Kong University of Science and Technology has already introduced AI-driven digital academics to deliver lectures in a Social Media for Creatives course, showcasing the potential of this technology to support and enhance traditional teaching methods. Strong Uncanny Valley vibes but still, wow…
Not entirely sure how to feel about this – both clear and obvious upsides (freeing up time for research and creative student support) and down (well, jobs). As this space continues to evolve, it seems essential for institutions to engage in open and constructive dialogue to ensure that AI is integrated in a way that enhances, rather than replaces, the invaluable contributions of faculty and graduate student instructors.
Sensemaking, AI, and Learning (SAIL) newsletter 👀
Looking for extra reading in the AI space? George Siemens (University of South Australia and President, Global Research Alliance for AI in Learning and Education) has a great newsletter that is very much worth a look – summarising the latest news in the space as it connects to AI, broader questions relating to the technolog(ies) and their implications, and meditations on where we are headed as it appends to education – “the fulcrum of learning has shifted. Knowing things will continue to matter less and less going forward as AI improves its capabilities. We’ll need to start intentionally developing broader and broader attributes of learners: metacognition, wellness, affect, social engagement, etc.” (Nov 4, 2023). Recommend – subscribe here.
“Explain it like I’m 5” – AI as an idea (and language) explainer
I’m a pretty simple person – maybe it’s a 🇳🇿 🥝 thing but I prefer plain, easy-to-understand language wherever possible – so one of the first use-cases I ever found for AI was to take complex ideas and get the AI to “explain it like I’m 5” (kind’ve a variation on the Feynman Technique, I guess). In AI workshops I’ve delivered over the last year or so, this has consistently been an extremely popular trick for a very wide range of different users.
I’m not alone here either – Kelly Webb-Davies (Bangor University) has a related take – regarding AI in HE assessment as a fantastic way to assist English as a Second Language learners in unpacking assessment tasks – using AI “to separate students’ ideas and knowledge from the language they use to express them”. Great use-case #recommend
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.
Have a great weekend ahead and let us know if there’s anything we’re missing that we should add to make this newsletter more useful for i) yourself and/or ii) others. This is a fast-moving, ever-evolving space and we greatly value any and all feedback. 🙏