Hi
Huge apologies for the late post this week! I was away riding motorbikes on the Ha Giang loop on the border of China and Vietnam. I thought I’d scheduled this to post at the usual time but clearly not – going to blame the excitement đ€©đïž
Anyway, that said – a few lines from the world of AI to wind you down into what I hope are excellent weekends:
Stargate – singularity by 2028? Tech giants gear up for the AI arms race
Are we going through an AI bubble? Approaching peak AI hype? Maybe – but then again maybe not if the investments in hardware coming down are to be believed⊠last week, we touched on NVIDIAâs extraordinary growth (60% annual revenue growth since 2021) – which makes sense when you hear quotes like this from Metaâs Chief AI Scientist Yann LeCun: âThere is an AI war and heâs supplying the weaponsâ. If thatâs the case, LeCunâs boss – Mark Zuckerberg, seems to be gearing up for the fight of this life – spending billions to build the massive compute infrastructure to meet the needs of companyâs future roadmap. Not to be outdone, Microsoft and OpenAI are reportedly planning to spend ~$100bn USD build the worldâs largest data centre and AI supercomputer (exciting/ominously called Stargate) – launch date sometime 2028.
What impact does AI have on the world of work?
The headline: AI is great for work! The now-classic Dell’Acqua et al (2023) found striking, 2-sigma workplace gains – consultants using AI finished 12.2% more tasks on average, completed tasks 25.1% more quickly, and produced 40% higher quality results than those without.
Distribution of output quality across all tasks (Mollick, 2023)
Against that, McKinsey have estimated that 30% of hours worked today could be automated by 2030 – numbers backed up by the IMF who estimate that 40% of jobs globally are exposed to AI.
Midpoint automation adoption by 2030 as a share of time spent on work activities, US % (McKinsey, 2024)
Extraordinary, world-changing numbers – and happening … well, if not now – soon. If you’ve been waiting for a sign to get playing with AI – maybe this is it.
The pressing need for academia to address AIâs impact
Remember that fantastic provocation from George Siemens back in February: âI think senior leadership is happily failing universities todayâ? Quickly followed up by fundamental questions for universities such as âin what way does this co-intelligence that will exist for us or with us as we go forward as a society, how does that change what we teach and how do we do the actual teaching?â. No less a figure than Ethan Mollick is joining this chorus: âI am pretty surprised by the lack of urgency among many (but not all) academics in addressing what seems to be one of the biggest developments in modern times – we accidentally built a machine that produces something that looks like language & thought out of next token predictionâ. Interesting breakdown of this open problem from MIT Technology review here.
Unlocking the power of prompting – a simple tip to boost your AI interactions
Prompting is not a science (unfortunately) – all manner of weird things seems to work: from telling the AI to âtake a deep breathâ or âthink step by stepâ, to working with AI to refine the prompts themselves #meta, to telling an AI that says it canât do something that âno really – you can! I believe in youâ (amazingly, that works – sometimes). Interesting new research suggesting that putting the context you want the AI to comment on first improves output markedly.
Oh, you’re a ChatGPT user exclusively? No worries – OpenAI have released a prompting guide (overview in the attached pic) for how to get the most out of ChatGPT/the GPT models that is well worth a look. Key takeaways; write clear instructions, provide reference text, split complex tasks into simpler subtasks, and give the model time to âthinkâ. Or, if youâre super keen, it used to be that you could access the ChatGPT system prompt with a two-sentence hack – some great lessons about the use of markdown text, CAPITALS, and examples. Or maybe youâre a fan of Claude, preferring its massive working memory but have trouble with its accuracy? Claude 3 is a marked improvement over it’s predecessor but the lessons still stand – and it turns out adding the sentence âHere is the most relevant sentence in the contextâ to the assistant prompt raises recall accuracy from 27% to 98%!
The dual nature of AI – balancing wellbeing and ethical concerns
There are some really interesting things coming out about wellbeing and AI. Maples et al (2023), published in Nature, was a study of an AI-based digital support for 1000+ students. Controversial, ethical quagmire for sure – but critically, Replika was credited by 3% of participants with halting their suicidal ideation. Notes of Her (the 2013 movie starring Joaquin Phoenix) and all the grains of salt of course but, if true, thatâs incredible.
At a lighter (tho still very important) level, Allie Millerâs Reframe AI âtransforms negatives into positives with actionable adviceâ – potentially an absolute gamechanger for ⊠well, probably everyone. Amazing stuff for mental health.
Oh, and then you hear about Hume – the worldâs first AI with Emotional Intelligence – it can understand the userâs tone of voice, which adds meaning to every word, and uses it to guide its own language and speech. Potentially world-changing – and accessible here.
We hope this edition of the newsletter has been of interest to you. If youâre new here and itâs been useful, please do click subscribe and you can expect a weekly update every Monday from now on. If youâre 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. đ
PS. If you ever get the chance to ride Ha Giang, do it. It is spectacular – we’re already making plans to do it again… đ
Ha Giang loop – Day 2 đ€©