I went for my daily walk today and I decided to listen to a podcast on my walk to stop myself from getting bored – I wore my wired earbuds rather than over the head earphones to avoid getting too hot. I decided to just do two laps and I am happy that I made that decision because I was exhausted after those two laps – a combination of heat and feeling a little dehydrated by the end of the walk but it was satisfying. I had a shower, got changed and then got myself prepared for the day.
I was listening to a recent ‘Windows Weekly’ podcast and then checked out the Microsoft presentation that was broadcast recently online with the launch of nVidia RTX Spark laptop and desktop. A few weeks ago there were articles talking about how Microsoft was downplaying the copilot branding as an indication of maybe Microsoft was scaling back its AI ambitions but I think it is a misreading of what Microsoft has decide to do. I think the problem with Microsoft’s fixation about AI is the fact that they did a lousy job showing to end users how AI integration into search can actually help make using their computer a lot easier.
In the case of using AI to improve the search functionality is the ability to type in something like “I want to change to dark mode” and the functionality appears in the actual search result where you can change in then and there. Compare that to the rather embarrassing use of AI a few months earlier, what they demonstrated was something that the average user finds useful. When Microsoft talked about integrating AI in tasteful ways this is what I am thinking – how can be we use AI so that AI isn’t the focal point but rather end users just recognise the fact that the search functionality got a whole lot better and not care about it was improved. A bit like the AI assisted photo editing, people don’t care about AI but they do care that there is a useful feature in their photo editing app that allows them to tweak a photo they’ve just taken.
The other interesting part of the presentation was the push into locally run models and models run on premises – there has been big push recently recently by nVidia and Palantir regarding ‘Sovereign AI’. As Alex Karp said, why would a business want to want to hand over their proprietary data that gives them a competitive edge – keep the model in house and protect your competitive edge. When it comes to local models running on local hardware then the benefit is that you don’t have to care about how many tokens you burn and if you do need to fall back to the cloud then that will be available. I think for many businesses who went into this whole ‘tokenmaxxing’ are going to start reevaluating their plans going forward – are they better off just buying each developer a decent workstation running an model locally with cloud based AI as a last resort?
If there is a move to more local models then the whole business case that justifies the hundreds of billions being spent on data centres is now undermined. If most if not all AI work eventually becomes something that can be run locally then where do these AI cloud providers sit in the equation? do they have a small cloud based business and charge a subscription for customers to download a local model they can run with the subscription paying for training runs, harness etc? If that is the future then it sounds like a move back from being a cloud provide to a software provider.
Apple released Safari Technology Preview 254 today and it includes a tonne of bug fixes and new features – I scroll straight to the section that deals with Web Extensions and it appears that it is getting some much needed attention. It’ll be interesting to see whether 27.3 includes many of these updates because I’m guessing 27.1 will be for new hardware support and 27.2 appears to be (based on the beta release notes) a pretty modest release rather than including some big updates.
