I’ve logged off from my work computer for the evening and this time I actually finished up at 10:00 pm rather than doing some extra work. I’ve finally cleared out the last remaining bit of work thanks to that extra work done after I clock off on previous days so now it is just a matter of maintaining it rather than having to make up for lost ground. It is one of the habits I picked up when I was working in fast food: a) always end the night and week with nothing required to be done for the next day so each day is start with a clean slate b) make sure you do a little bit of cleaning every day so then it doesn’t build up to become a massive job.
I was watching an interesting video a while ago and there was an interesting comparison between LLMs in the cloud being an unviable business model and then comparing it to rise in cloud gaming. I raise the comparison because when both LLMs and cloud gaming launched it was sold as ‘all you can eat for a flat price’ but now we’re starting to see usage limits, in the case of LLMs if you use a third party developer tool you have to pay the API price. What this demonstrates is that the business model isn’t viable because cloud gaming and LLMs aren’t like SaasS where each additional user has a negligible additional cost and even for a heavy user of a SaaS they will still only be using cents worth of resources when compared to the dollars charged each month.
How did they get the business model so wrong? because it was based on a set of assumptions that never made sense. The assumption was that it would be a matter of scaling up data centres and then eventually you reach a critical mass where your costs plateau as your user base increases. Basically it is the SaaS playbook but applied to cloud gaming but there is a small problem – the economics of SaaS cannot be applied to cloud gaming and LLMs because both of those services are so much more resource intensive. That doesn’t even touch on the fact that LLMs by their nature are probabilistic so not only are you going to have something expensive to run you then have customers slamming the system when the answer that is given is completely useless.
That doesn’t even touch on the issue of cloud gaming being a solution in search of a problem given how powerful smartphones are these days – even the Samsung A series has more than enough grunt for the casual gamer. Then there are game consoles, buy once and then buy your games – no usage limits on how much you can play the game for. Same can be said for LLMs – do you run them in the cloud and not know how much something will cost until it spits out a ‘kind of working’ solution or do you just buy a decent workstation from Apple or a Strix Halo AMD platform with 128GB of RAM then just the model locally – choose one of the free open weight models or an LLM from one of the big names like Google and others, tweak it and use it with a developer tool of choice.
Why do businesses do stupid stuff like the above? because they pander to idiot shareholders who don’t seem to understand that technology a) progress doesn’t occur in a linear upward fashion but rather there are sudden burst then a period of consolidation and refinement then out of no where another sudden burst of innovation b) the businesses that are household names such as Microsoft, Google, Meta, Amazon etc. are how mature businesses – the rapid hypergrowth days are behind them, they’ve now become a blue chip stock like IBM. Unfortunately these businesses appear to be going through the equivalent of a ‘mid life crisis’ and the equivalent of a middle aged man buying a sports car is a mature company throwing their money at something that they think will make them ‘cool, hip and into it’ to the idiots on Wall Street.
Matisyahu's Blog 