Matisyahu's Blog

The weekend has come to an end but looking back I’m pretty happy with how productive I have been over the couple of days off. I bought a new pair of shoes on Monday, picked up a television, sold off the Apple TV 4K and my old TV then purchases a Samsung smart watch. On Tuesday I went for my daily walk to give the new shoes a try – it felt as though I was walking on clouds the whole time. The experience was so much more enjoyable when compared to the hiking shoes I was wearing for the last week. I also wore my Samsung Watch and part of that has been learning how to customise the watch and navigating. When I first setup the watch i couldn’t work out why software wouldn’t update but then I found out that Wi-Fi isn’t enabled by default but when I enabled it i then found that there were a whole heap of software updates from both Samsung as well as through the Play Store that needed up dating.

When I went for my walk today I was able to kep track of my progress including my heart rate. Part of the monitoring was seeing how closely my estimated distance was when compared to what my smart watch says which I assume is based on GPS data. At the end of the walk i had done well over 10000 steps. I think having those statistics with me when I walk will keep me motivated along with being able to keep track of time when going for a walk. On Wednesday I’ll be getting up at 9:30am which gives me 15 minutes to get changed into my work out gear then get going for a walk – keeping track with the goal of getting it done in 1 hour and 20 or 30 minutes although funny enough I find that the walk in the morning tires me more than if I did the same walk in the evening.

Still a lot of hype regarding AI these days and the more I follow the development the more I see AI models eventually becoming commodified where the value won’t be in the LLM itself but instead who can deliver it competitively. What do I mean by that? those who end up winning run the model in a data centre with bespoke hardware so that the cost of delivering is done more efficiently than the current use of GPU to brute force their way through it (see how bitcoin mining went from GPUs to ASICs). Then once they get it doing the learning and inference as efficiently as possible (Google is already making big leaps forward with their own bespoke hardware) the next important factor is how it is integrated into their product portfolio – AI isn’t the end point but it is a means to an end.

Then there is providing it as a service to third parties through turn key solutions such as APIs that abstract the complexity so that third party developers can build their product utilising the cloud platform while making use of the underlying AI model. The other benefit of abstracting that complexity means that in the future it will allow the provider to upgrade the underlying LLMs with the products sitting ontop making use of it to inherent the benefits without having to write an additional lines of code.

What I do find funny are the number of executives who are quickly back peddling as they realise the whole ‘vibe coding’ ends up creating unmaintainable AI slop resulting in even more time and money being spent trying to clean up a pile of coherent barely legible code. It is always funny when I see executives get suckered into the sales pitch because rather than heading over to Amazon to buy a few books to understand ‘System Analysis and Design’ along with programming principles, they instead rely on consultants who have a vested interest to push whatever happens to be the product that their business partner is trying to push – see OpenAI and theim entering into multiyear partnerships with Accenture, Boston Consulting Group, Capgemini and McKinsey & Co (as reported over on CNBC (link))