Very few people are prospering off AI right now. We're in the part of the cycle where they're firing people to keep the stock going up.
Just wait until multiple companies start realizing AI might cost more to use than it is to hire people full time with benefits. It isn’t something that I think can take over as many jobs as the doomers think because just the electrical costs are absurdly high. Then you have to also train the remaining workers there to use the AI, which can sometimes be more redundant and actually take longer than doing it by hand. Then add in the fact that even with the advancements in AI, even for coding, you still have to understand coding enough to proofread the AI code to find mistakes.
AI does have its uses, but the grifters oversell its capabilities, while the CEOs view it as an instant profit increaser. It’s then this endless feedback loop between the AI grifters like Sam Altman and the CEOs like Bezos trying to make the most money possible through sheer hype alone, which typically ends in disaster if said product fails to live up to that hype.
The most frustrating part of the whole "AI" debate is that a legitimately useful technology was severely overshadowed by all the goddamn marketing hype and legitimately predatory practices. Ironically, OpenAI squandered its considerable lead in the market by allowing Sam Altman back into the company in 2023. Sam Altman isn't a Steve Wozniak where he knows the technology inside and out and can't try to talk it up in all good conscience. He also ain't (despite posturing like) Steve Jobs, in that he's a ruthless businessman with ironclad fixed principles. No, he's something far worse: a gutless, spineless coward who talks a big game like Sam Bankman-Fried and Braden Karony, but his only real marks are stupid fucking boomers who can't process technology newer than Windows 3.1.
a) Claude, created by Anthropic, itself formed by disgruntled OpenAI employees. Claude, regardless of model and version (i.e. Haiku, Sonnet, Opus),
blows current GPT-5 even past GPT-3 and GPT-4 models out of the fucking water. Imagine getting mogged hardcore by your ex-employees who
don't have a hypervisible AI hypeman like Sam Altman at the forefront. That's gotta be embarrassing.
b) Google was well and truly behind the curve for like a year and a half when OpenAI partnered with Microsoft and actually made Bing useful for that brief stretch of time. Then Gemini happened and gave us an alternative to OpenAI hegemony be it directly through GPT models or indirectly through Claude.
c) Bing AI was genuinely fucking excellent stuff on launch. It was
supplementary, not
primary like Copilot is. Unfortunately, Satya Nadella just
had to double, triple, quadruple, and quintuple down on shitty fucking business decisions and turn an excellent tool confined to Bing into a broader invasive ecosystem through Copilot... and he didn't even have the decency to revive the Cortana branding with the OpenAI partnership. NO ONE WANTS AGENTIC ANYTHING. WE JUST WANT TOOLS THAT WORK AND STAY OUTTA OUR WAY.
d) Mark Zuckerberg was snorting an assload of cocaine while shooting up ketamine during the pandemic lockdowns and seriously thought the Metaverse was where the money's at. When that went tits-up and his investors were writing sternly worded open letters, he decided "okay, well, plan B" and decided to
flood the market with FOSS models like LLaMA with transparent data regarding training weights and everything. Ironically, Mark Zuckerberg gave us one of the best self-hostable LLMs out there in LLaMA... and there's literally no fucking way for Meta to harvest my data off my local Ollama instances.
e) The Chinese needed to make us white monkeys quiver in our boots, so they created DeepSeek and Qwen, both of which (DeepSeek especially) proved to be remarkably power efficient relative to the bloated GPT models that were around on launch. Even if DeepSeek and Qwen were bootstrapped with Western technology that they misappropriated, the data don't lie. These models are demonstrably more efficient while achieving broadly similar outcomes. Qwen and DeepSeek are
frequently among the most popular self-hostable LLMs on offer.
f) Of course, none of these happenings matter in the slightest because the average normie on social media hates AI for invasive data harvesting (agreed), theft of copyrighted material (agreed on a technicality because of the lawsuits but definitely understandable), data centres consuming huge amounts of energy and fresh water resources without meaningfully contributing to the locales they operate within (definitely agreed), etc etc. Unfortunately, both Biden and Trump 47 have done
jack fucking shit to keep these companies in line because "GDP LINE GO UP."
That's not even getting into all the obscure developments and differences that happened the moment that ChatGPT and Stable Diffusion were made mass market in late 2022/early 2023. AI derangement is a bitch and a half to overcome, and so many people are basically digging their heels
to their own detriment. I also fucking hate all these platforms like
Hugging Face and
Ollama outwardly present themselves as "easy" or "user friendly" or demystifying these seemingly obtuse black boxes called LLMs... and that goes right out the fucking window when you're actually perusing the sites and then you get lost in technobabble, parameter counts, and vague descriptions that don't actually tell me anything useful. Oh and don't you
dare expect the fucking documentation for
either of them to be up-to-snuff. NOPE. It's just "oi m8 here's how u bootstrap Ollama and pull a model. GUI? What GUI? Ur on ur own m8."
A local LLM won't ever fucking resemble GPT-4o, Claude Sonnet 4.6, Gemini 3.whatever, etc etc. That's just a fact of life. Y'know what you can get away with running a basic bitch 7B-14B parameter self-hosted LLM like LLaMA 3.1, Qwen2.5, or DeepSeek-R1? Translating stuff, sanity checking shell scripts (important for me because I'm a turbosped who daily drives Linux), actually giving you boilerplate templates for stuff like cover letters where you
could be original but this is for a job application you're sending out by the dozen anyway, y'know... stuff along those lines. It ain't
much, but local LLMs can also fetch automatically generated transcripts for protracted Teams meetings, and then enumerate all the new items vs old talking points vs items that haven't been actioned yet despite repeated attempts to highlight it. You don't
need to pay out the ass for Claude or GPT tokens when you already have good enough local, wholly self-hostable LLMs that you can run on bare metal already in the company server closet, let alone your personal gaming rig from 2017-2020.
Sadly no. mass market LLMs will forever be tainted by the inexorable, miasmic odour that Sam Altman proliferated throughout the entire industry and popular perception. Oh yeah, big companies on the whole (including NVIDIA, Oracle, RAM manufacturers like Micron and Samsung, and blah blah blah) had a huge hand in souring our opinions on AI. But I specifically credit
him with giving us a face to pin on the sordid status quo that we've had to live under for the last 3.5-4 years
even if blaming him wholesale is technically unfair.
