Hi
Hope you had great weekends wherever you are – things here in Hanoi are getting increasingly dragon-themed as preparations for Tet start gathering pace (12 more days!). A few headlines from the world of AI to kick off your Mondays:
Deep fakes are here for the public sphere and elections – “truth is the first casualty”
You’ve probably heard about the deepfake pornographic images of Taylor Swift on X and Telegram – viewed tens of millions of times before being taken down. Shocking and perhaps the start of a worrying trend of increasingly widespread deepfakes in 2024 – which includes, in the same week, a Joe Biden AI deepfake telling Democrats not to vote. Not perfect but very believable and concerning given similar tools for celebrity voice deepfakes are widespread or easily made using tools like ElevenLabs. This in turn raises real questions: are our systems ready for AI at this scale?
Because the potential is broad – celebrities, public figures … even members of family being impersonated, with personalised mass phishing calls, fake news and more… There are slight signs of hope though – guardrails are being built into tools (see the pic below when I asked for a picture of Donald Trump – though easily circumvented – see the banner image at top), there are suggestions that unreal light sources and other elements might be a tell that can be used to detect AI-generated images (for now, at least). And, in welcome news, it seems like the US might be joining the UK in banning the creation and sharing of deepfakes. Looks like we’re in for a bumpy ride here…
Emerging image generation guidelines from Midjourney
OpenAI enters academia by partnering with Arizona State University
It’s happened. ASU will reportedly have exclusive access to ChatGPT Enterprise from next month (February), and plans to deploy the AI system across coursework, tutoring, and research (though no student access at this point, apparently). This engagement with AI is not surprising in and of itself – universities from the UK’s Russell group to Australia’s Group of Eight have committed to doing this – but this is a new power move from the ChatGPT maker itself.
Interesting times – you wonder how Microsoft feel about it with the roll-out of the genuinely gamechanging MS Copilot (think AI in Word, Powerpoint, Excel, everything M365 – available in Aus now and reportedly coming to Vietnam “soon”) – though with a $30USD per user per month price-tag for Copilot, maybe ChatGPT for Teams or ChatGPT Enterprise is a better offer…? 🤷
New study shows AI can improve mental health outcomes
Student mental health is a crisis for students globally with analyses showing “alarming” figures that only worsened with the COVID pandemic. A recent study published in the Mental Health Research journal focussed on mental health and, particularly, loneliness and suicide prevention using a GPT-3 enabled chatbot.
In a study of 1006 students, 90% of participants noted in an initial survey that they experienced loneliness – with 43% being Severely or Very Severely Lonely. Over the course of the study, significant benefits were found: 23.6% of participants experienced positive life changes (e.g., mood, behaviour, or overall wellbeing) ; 18.1% experienced therapeutic results (i.e., benefits in dealing with mental health issues and/or emotional support); and 3% reported that their suicidal actions were prevented because of their engagement with the chatbot. That last by itself is extraordinary – those are lives literally saved directly because of AI.
The twist? The chatbot was Replika. An AI “for anyone who wants a friend with no judgement, drama, or social anxiety involved”, Replika has been the source of some controversy in the past (think Samantha from “Her”) but maybe that last needs revisiting under certain circumstances. After all, for some, it could literally be a lifesaver.
Spare a thought for Google
Because they are really starting to heat things up:
- Gemini Ultra is coming soon and the suggestion is it might just be better than GPT4. And that’s saying something really as GPT4 was released mid-March 2023 – wild it’s stayed largely on top of the pile since.
- A new model called Alpha Geometry has just made headlines with its ability to solve wildly complex maths problems at the level of a human Olympiad gold medalist (apparently an extremely high-level high school maths competition). Able to do this without human demonstrations (here in Nature or a less-technical overview from Deepmind here or NYT (paywall) here this is a concrete step forward in both logic and creativity by AI.
- Google is making power moves in the education space, AI Track of the Google for Education Pilot Programme – reportedly this includes a lesson planning copilot, lesson sets of personalised activities for students with adaptive, AI-powered feedback and support, and a roadmap that features “help with evaluating student work”. Runs off of Google Classroom so might not be available for your institution but you can check out an overview of the offering here.
Lumiere = new State of the Art text-to-video generator? And text-to-music with Google Labs
More Google! There might just be a new best-in-business AI video generator. Google’s Lumiere looks amazing (see below) and user evaluation tests suggest people prefer its output to Runway, Pika Labs, and Stable Diffusion Video. Looks like there might be a lot of advantages with video in-painting, image-to-video, and stylised video generation.
Lumiere demo (Google Research)
And if you want to get a bit more out there, check out Google Labs – home to their latest experimental tools and technology. You might need a VPN but things like MusicFX (strapline: “describe a musical idea and hear it come to life”) are pretty interesting time-sinks. Check out “music for a coffee shop in HCMC when the streets outside are flooded because its rainy season and its pouring but that’s okay because you’re warm and dry inside with a ca phe and banh mi on its way” – still muzak but definitely interesting.
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 an excellent week and catch you next time!