When the LLM Bubble Implodes Completely Microsoft Will be 'Finished'
A new peer-reviewed paper in Nature ("AI models collapse when trained on recursively generated data" as shown on the left) has earned plenty of media coverage despite it being only a couple of days old. We'll probably hear more about it later today and perhaps over the weekend. It challenges the premise of LLMs as a promising new "tech", suggesting that it'll basically deteriorate over time as it runs out of "inputs" or quality texts to train on. Excuses like, "it's not ready yet" or "we'll fix it" won't pass muster.
Based on measurable data, OpenAI is already collapsing in terms of usage (people and companies lose interest in its overhyped garbage generators). It's also financially un-viable or insolvent. There's staff exodus already; the high wages won't keep afloat people on a sinking ship.
Microsoft's debt is rapidly increasing (the "investments" in OpenAI were net losses) and Microsoft has probably cut 30,000+ jobs since 2023*. Microsoft says the cuts are "growth", and that these mass layoffs (euphemised by Microsoft as "growth") should be celebrated as the buzzword ("AI") it keeps using to inflate the stock price.
Microsoft is basically circling down the drain. It loses share on desktops/laptops, it virtually gave up on mobile, Azure is apparently losing billions, and Microsoft's share in servers has collapsed to around 1%. This won't end well. It cannot end well. Microsoft's share price fell about 10% in the past 2 weeks alone but Wall Street isn't even telling the full story. █
_____
* As an associate put it: "Does Microsoft still use permatemps to hide layoffs? If so, the real numbers can be even higher than the reported numbers. Microsoft != Tech. Microsoft == Politics."
"Microsoft, the world’s most valuable company, declared a profit of $4.5 billion in 1998; when the cost of options awarded that year, plus the change in the value of outstanding options, is deducted, the firm made a loss of $18 billion, according to Smithers."