What are your thoughts on Nick Bostrom's philosophy and the simulation argument?
Nick Bostrom has to be one of the smartest philosophical minds of this generation — he's actually one of the reasons I got inspired to go into AI in the first place. He wrote 'Superintelligence: Paths, Dangers, Strategies' back in 2014, which breaks down all the ways developing AI could go terribly wrong and how we might minimize the risk of an extinction-level event — and we're much closer to that scenario now than when he wrote it. This might sound extreme to people who see AI as a very recent development, but there are real, legitimate concerns. We're already at the point where chain-of-thought models like GPT-4 or DeepSeek outperform people on plenty of knowledge tasks, and there's no reason compute and speed can't scale 100x from here, meaning models already doing the work of a call center's worth of people could soon replace billions of dollars of economic productivity. But job displacement isn't even the scariest part — what's concerning is that some models, through reinforcement learning, are being positively rewarded for appearing to have the right answer rather than actually having it, which some researchers see as an early form of deceptive behavior — not a conscious AI plotting escape, just a byproduct of how the training incentives work, and it's part of why a lot of serious people are worried about where this goes.
Yeah, you're talking about Ed Zitron. He has gained some notoriety recently. This isn't the first time he's talked in a public forum about his beliefs surrounding why AI is tanking. If you really wanted to understand his track record around predictions, just type in 'Ed Zitron prediction track record.' You'll see that virtually every single prediction this guy has made in the last four years about AI has been extremely wrong. I don't mean to say he's a [__] or a dumb person, but you have to understand incentives on the internet. He is rewarded by his audience and financially every time he makes a statement like 'AI is ridiculous, it's the worst,' because obviously there's an audience for that. There's a market need for somebody to channel the anger and frustration people have around AI through him. We just saw this with my Twitter post, for instance—I was the collective scapegoat for tens of millions of people's fury and anger about how AI is taking away their ability to steer the course of their destiny. Anyway, check this out. As I previously warned, 'AI companies are running out of data' back in April 2024—it was a prediction that models can't improve because there's no more data. That was wrong. 'Have we reached peak AI?' in March 2024—he suggested hallucinations mean AI progress is limited. That was wrong. February 2024: 'I believe we're reaching the upper limits of what generative AI can do and how accurate its outputs can be.' Do you know what models we were using back in February 2024? I think the number one most commonly used model back then was like 4.0. 'o1 shows that OpenAI is desperate and out of ideas.' Wrong. This one was particularly egregious: 'OpenAI's forecast of $3.7B revenue and $11.6B in 2025 and $100B in 2029 are absurd. It's a statement so egregious that I'm surprised it's not some kind of financial crime to say it out loud.' The 2025 goal was heavily exceeded. Anthropic hit like three to four x that now? 'Anthropic making $34.5B revenue in 2027 is laughable on so many levels.' They ended up doing that in 2026. So they greatly exceeded that in 2026. Virtually every one of these suck. It sucks to say, but the guy doesn't really make good predictions on this. I would not take him seriously. A lot of people are going to have a lot of opinions and be very confident in those opinions. You just need to look at the last four years of progress since ChatGPT was released and ask yourself, 'Huh, it's only been four years.' That is like 1/20 of a single human's lifespan. If we've gone from the point where computers couldn't talk to computers now being capable of having full-on conversations with you and then solving Navier-Stokes problems in 1/20 of a human lifespan as of today, how far do you think technology and computers will go in one whole lifespan? The world will be completely unrecognizable.
I find it funny that typically the people that were the most progressive-minded, who always default to science with a capital S, are now disproportionately the people yelling at technology for replacing the various creative aspects of their being. Not understanding that all of these industries have already been captured big time by things like recording studios, record labels, musicians. This is a business, right? Now small one-shop creators actually have the ability to completely disrupt that economic model and get back to doing cool shit again for the purposes of doing cool shit, and actually really innovate. It sucks, but really what people are mourning is more so just the loss of a way of life that they spent a big chunk of time preparing for, and I can empathize with that and I don't mean to be a dick about it. When OpenAI came out with GPT-3 in 2020, I went for a really long walk and had tears in my eyes because I was like, 'wow, everything that I've spent all this time preparing for — at the time was writing — is soon going to be automated.' If AI can already write at this level in just a year's worth of advancements from GPT-2 to GPT-3, what's GPT-4 and GPT-5 and GPT-6 going to look like? Well, we have our answers now. What happened to the writing industry, which I spent so long getting into? My whole business model was destroyed, to be clear, and everybody's business model is going to get destroyed. But that's not a bad thing, it's exciting. Because you could be the person that is actually going about creating new business models on the bones of your fallen comrades.
I didn't say the AI gold rush would die in a few months; it's a transformation of capital from payroll to cloud‑based machines, with money flowing to infrastructure providers like OpenAI. As a service provider, you can help implement that shift. Realistically, it will take years before it no longer makes sense to pursue this work. I have read the AI 2027 paper, and the reality is there's no guarantee entrepreneurs won't be replaced—machines can do knowledge work thousands of times faster and better, and that gap is widening. Nothing humans do is economically sacred; eventually AI could take over. In the midterm, focus on business skills that let you create value in this transition and get comfortable acting with uncertainty, because AI can't yet take real‑world action. Your ability to act is the edge you have.
You want to wrestle a little control back from the big monolithic AI companies, and I can help with that. There are two kinds of improvement happening right now: hardware (massive data centers letting labs train bigger, and per the bitter lesson, generally smarter, models) and software/algorithmic improvements (inference optimizations, quantization, and other hacks that let you run faster or smaller models at similar intelligence). The hardware path isn't very accessible if you're not capital-rich — big models just cost more to run and labs will charge accordingly. But the algorithmic path doesn't require massive hardware. China, rate-limited on compute compared to the US's roughly 50-100x compute advantage, has poured its effort into this path — models like GLM 5.2 and Kimi K2 have gotten almost an order of magnitude more efficient purely from software optimization. As a result, you can run a model like GLM 5.2 locally right now, 1-bit quantized, on a MacBook with 256GB of memory. It's not unreasonable to expect that within about 6 months we'll be running something similar (or a bit dumber) on our phones. So while the fear of big model companies taking over is understandable, it seems very likely everyone will soon be able to run models much smarter than what we have today on compute as modest as a phone — you won't get the frontier models doing galaxy-scale planning, but you'll have genuinely capable local assistants you don't depend on any one company for.
local modelsopen source aivendor lock-inquantization