I woke up this morning at 9:30 am and then got out of bed – although I didn’t have a long sleep last night I did want to stick to my goal of getting back into a good routine. I got my house organised, had a shower then walked down to the local bakery to have brunch and while eating brunch I watched a few YouTube videos on my phone. After finishing brunch I bought some items from the display cabinet for dinner tonight and walked back home – not much of a walk but a walk nonetheless.
I was watching a video regarding the backlash against technology – in particular the technology hype around AI:
For me technology just straight up sucks these days for two reasons – the basics aren’t being done well resulting in a crappy experience over all then to compound that further you’re now subjected to AI being rammed into every nook and cranny with no discernible benefit to the end user. When it comes to AI integration, in many cases it is difficult to avoid it unless you’re willing to go out of your way to hack the software to remove or disable such functionality.
For me what symbolises the ‘technology sucks today’ narrative is how the quality of software keeps going from bad to worse – I’m not asking for new features but rather making sure the basics are done and done well. I’ve routinely complained on this blog about Safari and it’s half baked attempt as supporting web extensions, many have mentioned the half baked delivery of liquid glass with the launch of version 26 of their platform. In the case of Windows 11 I routinely hear complaints about Microsoft ramming AI into every nook and cranny while they cannot be bothered to fix up their UI so then dark mode works properly, finishing off migrating control panel compoents to settings ever since moving to the new UI which started back in Windows 10.
Unfortunately what we see with the AI hype is the product of capitalism – rather than these technology companies accepting that they’re now a maturing part of the economy they’re desperately trying to convince themselves that they’re still hip, cool and in with it like an old guy convincing himself he still has it. The other part of the equation is the fact that venture capitalists have been producing negative returns for quite some time because the big hyper growth opportunities are few and far between. If these venture capitalists were honest they would hand back the money to their investors and close up shop but unfortunately venture capitalists are like a social movement that reaches its goal and is now trying to justify why it should continue existing.
The era of technological leaps forward has come to an end at least for now so the focus should be, by the technology sector, to consolidate the benefits of what they already have and make them work better rather than looking for a new distraction. Can LLMs and AI in general be beneficial? sure but I don’t think it is a multi trillion dollar industry. I think there is opportunity for the models improve from both a hardware demand as well as accuracy but what I think we’ll see long term is a move to locally run models. Developers will start preferring to use specialised models or open weight models which allow them to tweak the model for a particular task. When you own the hardware and run the model locally then you’re in control – no guesswork over how many tokens you’ve used.
Apple has done a good amount of work with their own SoC but AMD is rumoured to have Zen 6 SoC with a built in NPU along with the purchase of Taalas which could mean a PCIe accelerator card where the silicon is designed from the ground up to run specific model. How fast? Taalas can run the Llama 3.1 model at 17k tokens per second when compared to an nVidia B200 doing 354 tokens per second. Intel is also doing their own work in the field of NPUs and GPUs along with work being done by Microsoft and others in terms of getting AI models running on CPUs without any loss of precision.
Getting back to hating the tech dudebros and the whole AI hype cycle. I think the biggest worry people have is how connected the AI industry is politically and when it all turns to crap then people are concerned that it’ll get bailed out. This is the reason why I look at the recent proposal of a $1.5 trillion budget – not just once off funding for build up armaments and equipment but a shift to a defence budget of $1.5 trillion. For that amount I could imagine if Oracle finds themselves in a spot of bother I wouldn’t be surprised if we end up with the Pentagon claiming that these AI data centre projects are of national strategic importance and nationalise them which will bail out Oracle. Don’t be surprised if we end up seeing Palantir Sovereign AI running on said data centres.
When I went to Ireland on my trip overseas the best meal I had was at a local restaurant where I had seafood chowder with fresh crusty bread and butter. It was a rather overcast and miserable day but the meal offered something comforting which made up for the lousy weather.
Matisyahu's Blog 
Leave a comment