As of the time of writing, even the mainstream financial press is beginning to doubt whether the large language model market (what we laughingly refer to as AI) is sustainable. Even the gaffer of the Bank of England has gone on record to register some doubt, albeit couched in the wildly conservative words that comprise English understatement. Our colonials might translate his utterances as “FUCK, run for the hills, boys!”
I’ve been thinking about the next five years, and how they might play out in the world of technology. The financial implications of the LLM bubble bursting are well outside of any of area of expertise I might pretend to have, so I’ll park those and start buying dried pulses and bottled water instead of pontificating. But tech? I can have a shot at predictions in that sphere. Thus this thought experiment. Bear with.
LLMs go tits up
Inevitable. They’re not good at 99% of the things they claim to be revolutionary in, they cost too much to run and train, and the whole shebang is propped up by debt with no financial solution in sight. This will not continue. Sorry about that.
All change in Silicon Valley
Or rather, all shuffle around a bit in Silicon Valley. From what I understand: Oracle goes away permanently. Meta takes a big hit (maybe with that Mark Zuckerberg shoved off into retirement). Google absorbs Anthropic. M’Soft absorbs OpenAI.
AI is scaled back, data centre deals go to the wall, retreat, retreat, retreat. Google Gemini-powered search is tweaked to serve ads (it’s Google’s natural stamping ground), Gemini disappears from Google Docs and the like, Copilot disappears unless you have a Business Dickhead-Tier Licence, and everything is neatly compartmentalised and rolled back.
No one dies, no one takes a fall
There is no one to blame for any of this. No one from a tech company board goes to prison, governments don’t impose meaningful regulation to stop it happening again (see: every financial crisis since the beginning of time). Journalists publish articles titled “What went wrong with AI?” as if they knew what was happening all along but were forced to reprint OpenAI press releases like they were the Gospel.
Now, imagine it’s 2031:
Only the rich can afford to use AI
Given that the cost per token for frontier models is likely to go up somewhere between 6x (maybe break even) and 10x (make reparations to the creditors) or even 12x (and make some money on top, too) the only users of large language models will be those with deep pockets. Given that this shit is going to be expensive, it had better work. Oh, bugger, there’s a few issues there. Hang on, what can large language models actually do?
LLMs are good at code and computer stuff
Sort of, as long as you keep an eye on them and know your onions already. So big companies can equip some software engineers with LLM code assistants. Cybersecurity issues are discovered by LLMs run by a smaller, more centralised cabal of big tech companies who issue CVEs like a Roman Emperor making pronouncements and the rest of the world scuttle around trying to fix stuff.
Chatbots are gooderer with an LLM, but have only three suppliers
The “best chatbots” (oxymoron, ahoy!) are only available from M’Soft, Google and maybe Meta. This will have the lovely side-effect of a Google/Meta/M’Soft Chatbot embedded into the infrastructure of every company, thus supplying lots of lovely data to Google/Meta/M’Soft, so they can sell more ads or data to companies that sell ads.
AI social media creation
AI-created social media hits new creative lows, with “content creators” using the cheapest models they can to produce more episodes of Celebrity Fruit Snack Love Island for a drooling populace.
Any other use that may benefit humanity
LLMs will be available to your local school, hospital, police force, fire service or government at eye-watering prices from big tech companies who will charge a premium for the privilege. Public funds increasingly siphoned off to either Silicon Valley or Beijing.
But it’s not all bad news!
Brain rot starts to decline
People have to find things out, read things, and make sure they’re true. Writing, painting pictures and making films & games become valued human pastimes. The average IQ of humans increases. Everyone looks back and reads articles with “Where it all went wrong” in their titles.
Everyone stops talking about AI
Nuff said.
Everyone starts talking about why it’s getting so hot
Yes, why is it so hot? And windy? And on fire?