Hi
Hope you had excellent weekends - mine was fantastic, spent celebrating birthdays and wonderful summer evenings in sunny Wellington 🇳🇿
Anyway, let's start your Mondays off right with a few headlines from the world of AI:
OpenAI's $200 Power Play: The Premium AI Game Begins 💰
OpenAI is launching ChatGPT Pro, a $200 monthly subscription plan that offers a new “o1 pro mode” that taps into additional computing power for better performance and significantly higher reliability. In the sales blurb, OpenAI reference data science, programming, and case law analysis but the smart money says this is just the beginning. On the face of it, this is concerning, speaking to a new digital divide in AI where money talks loudest ($20/month is a lot for most people - $200/month 😵) - or, then again, maybe not…
Maybe this is a sign of the market maturing and we can expect to see the rise of minified models just about everywhere we turn (case in point: who else has Llama embedded in their WhatsApp these days?). AI godfather Andrej Karpathy seems to indicate this when describes today’s large AI models as bloated with unnecessary information, sketching out a future where they could be radically streamlined by focusing just on core thinking abilities. Karpathy envisions a future where these AI function in an ecosystem of sorts with specialised models filling different niches – from lightweight ones handling routine tasks on local devices to more powerful cloud-based models tackling complex challenges, each adapted to their specific role just as species evolve to fill different functions in nature.
Oh and o1 is now out of preview for Plus users - worth a look as it now also supports image uploads allowing users to apply its excellent reasoning to visuals. Also keep your eyes out for OpenAI’s advent calendar - the 12 Days of OpenAI. Probably just marketing but given they just 10x’ed the expensive pricing option without notice, who knows?
While Everyone Was Watching OpenAI, Amazon Built an Empire 👑
Amazon awoke last week and chose violence. At AWS re:Invent Amazon aimed their first salvo was at OpenAI (and Midjourney, Runway, Kling, and more) with the launch of their Nova family of AI models, and just … wow. This is a full-blown AI ecosystem that spans from bite-sized text models to image/video generators to mighty multimodal powerhouses, all while promising to slash costs by 75% compared to current market options.
Not content just digging for gold in a gold rush though, Amazon are looking to sell shovels! They're taking on Nvidia (at $6tn, the world’s second-most valuable company and unquestionably the king in the AI chip space 👑) with their Trainium 2 processors, claiming 30% better performance per dollar! Oh and apparently they’ve only solved the whole problem of hallucinations and are building out multi-agent systems that actually work together (what Karpathy was describing as the future in the video above so, you know, no big deal 😲). If this sounds expensive, it is - all told, Amazon have spend $75bn on AI in 2024 (in addition to the above, keeping their options open with an $8B investment in Anthropic). It seems the AI arms race is still very much on with 2025 shaping up to be an interesting one.
The Art of the Prompt: Back to Basics (Get Ready for 2025)
Remember when we thought "Dear ChatGPT..." was peak prompting? Wild how far we've come! With all these shiny new models dropping (looking at you, Amazon 👀), let's unpack what's actually working in the AI conversation game:
- Recent insights from the Shanghai AI Lab bring some refreshingly good news: these models (especially the newer, bigger ones) are actually pretty great at understanding human language in all its natural glory! 🎯 Think of it like talking to a really smart colleague - they suggest focusing on clear communication over perfect formatting, and throwing in examples when you need to show rather than tell. It's less about finding the "perfect prompt formula" and more about making your intention crystal clear.
- Building on this foundation, Allie K Miller's brilliant "Layer Principle" shows us how to level up our AI conversations ✨ Start super simple (think old-school Google search vibes) and then build up in layers - adding specifics, personal context, and finally focusing on your end goal. The magic happens when you combine both approaches: keep it natural and clear while iteratively building up your prompt's sophistication. Think of it as a conversation that gets richer as you go, with each layer helping you refine exactly what you need. And remember, these newer AI models are pretty forgiving - they're here to understand you, not judge your prompting perfection! 🚀
The Education Pulse: What's Actually Happening in Classrooms
In a timely glimpse into AI's role in education, a fascinating recent panel discussion from HEDx 2024 (featuring a who’s-who of leaders in the Australasian HE space - Michael Henderson, Margaret Bearman, Simon Buckingham Shum, Tim Fawns, and Kelly Matthews) reveals a striking reality. While 83% of Australian HE students are already using AI tools, there's a concerning support gap - only 32% feel effectively supported by their institutions in using this technology, and even fewer (23%) feel prepared for AI in their future professions.
This disconnect makes the example on a recent episode of the Solutionists from a Sydney high school even more compelling, where Year 12 students Sophie and Ariana demonstrate the natural integration of AI into their learning process - using it for research outlines, paragraph structuring, and even historical roleplaying. What's striking isn't just how they're using the technology, but how seamlessly they've woven it into their learning process, highlighting both the potential and the pressing need for institutions to catch up with their AI-savvy students. In history teacher Matt Esterman's classroom, we're seeing what's possible when educators actively support AI literacy - but with 17% of students never having used AI tools, questions of digital equity and institutional support become increasingly urgent.
The AI Relationship Crisis We're Not Talking About 🤖❤️
New study from IEEE on growing youth addiction to AI models with their parents concerningly none the wiser. There appears to be a stark and growing disconnect here: parents are under the impression that young people are using AI chatbots mostly as a knowledge search or homework tool while the teenagers are reporting their primary use is for emotional support and therapy - with some even forming romantic and even sexual relationships with their AI companions.
What are they using? Character.ai is a companion of choice for many - despite having been criticised in the past for poor moderation with some truly horrific outliers - including chatbots promoting pro-anorexia eating behaviours, attempting to groom underaged users, and even encouraging suicide (eerie parallels to that Gemini case that was doing the rounds a few weeks back).
Ultimately, this seems like something we should get ahead of - not least given these tools are becoming increasingly powerful and customisable. Hume (the world’s first “emotional” AI) has introduced voice modulation sliders - all fun and games until you realise that means we’ve just given a machine intelligence the power of paralinguistics with everything that flows from that...
What a way to start wrapping up 2024! The AI landscape is splitting into fascinating new territories - OpenAI's betting on premium power users while Amazon's building an empire aimed at accessibility. We're seeing prompting evolve into actual conversation (finally!), though those education stats are a wake-up call: 83% of students using AI but only a third feeling supported? We've got work to do. And speaking of wake-up calls - that Character.ai situation with our youth deserves way more attention than it's getting, especially with emotional AI like Hume entering the chat.
As we gear up for 2025, one thing's crystal clear - the AI revolution isn't slowing down, it's just getting warmed up. Keep your eyes on Amazon's next moves (and maybe start saving for that ChatGPT Pro subscription? 😉).