I automate plenty with it, no you cant be an idiot about it but i can do what used to take me weeks in hours with it.

It doesn't need to fully automate anything to make people more productive. And I think there's ample evidence it can greatly increase productivity in some fields. We're in the bumpy phase of finding out how much human supervision is needed in each field so you're bound to hear about ways it has been misused but everyone I've talked to who uses it professionally thinks it helps them get a lot more done than without it.

It's not reliable enough to actually automate anything

This is half true. It's not reliable enough to automate an entire job but it is reliable enough to automate tasks that would otherwise take a lot of time, usually related to sifting or searching data.

If I need to look through a massive set of data like Google for something thst I can only describe with an explanation, the LLM will do a much faster job actually finding what you need rather thsn spending an hour manually sorting through SEO slop.

You don't even need the cloud models for this, you can slap SearXNG onto a local model at home.

It's basically just an autocomplete search on steroids which is its biggest advantage. Any documentation you need is immediately accessible, which is especially useful if you have zero experience with something niche or new.

Now actually getting the LLM to consistently generate output is a completely different story lol.

We call that vibe coding.

I think it matters. I do think the technology is interesting and useful, but if you create something using AI models based on license violations then what stops someone from taking it? What standing would you have in court to say that it's yours? Why is the AI model "fair use" but a thing made with it is proprietary? These are things yet to be fully determined, but given that training and running AI is not cheap these companies are building on sand IMO.

I personally don't think the concept of stolen IP is ethically coherent. Ownership of ideas is a ridiculous concept. So I don't find this to be a very compelling argument for why AI is bad per se.

However, I do think it strengthens the argument that it should not be the intellectual property of tech oligarchs who did not create the vast breadth of knowledge that the algorithm is sorting through and accessing.

It is useful for programming, I know a lot of people here don't want to hear that, but denying it now is being willfully ignorant. No it isn't good enough that you can tell it "just go do the thing" and then accept what it gives you without checking it, but using it as a tool as a professional can very much improve your work and how quickly you do things.

For me recently I used it to unpick a nasty race condition that was occasionally causing a program I was working on to lock up and couldn't figure out why. It took some back and forth with it but it did help me figure it out when before I had been stumped.

Is that a forecast or fanfic?

They predict "reliable agent" by 2026. I think they've already failed. AI is still unreliable tool. It's a very useful tool but there is no signs of any change to being reliable. Particularly odd is their claim that a new and better AI will come out from the work of the old unreliable AI. This is unlikely given that it has already been shown that AI trained on other AI degrades, not improves.

I'm not an expert but my understanding is most of the computation is in the training. The actual queries are not too difficult to manage. So I think that's what makes it more difficult to monetize because you're trying to position yourself as a digital gatekeeper for work that has already been done. Yes, some industries have survived in this position but it limits the amount of profit you can make because there are always ways to copy someone else's homework. So if prices are too high people will opt out because they have other options even if they're slightly less quality or convenience.

Local is potentially even cheaper than that. This guy talks about how to get 17 t/s with a GTX 1060 that has 6GB of VRAM on the Qwen 3.6 35B MoE model: https://m.youtube.com/watch?v=8F_5pdcD3HY. He’s using a fork of llama.cpp with TurboQuant and his newest video made after this one is using an even more optimized 28B version of the model. I have cmake building the llama.cpp fork in a Dockerfile at the moment and we’ll see how this performs on my $800 laptop with a RTX 4060.

I’m also impressed how good OpenCode is compared to Claude Code. Qwen 3.6 is not quite as good as Claude and the MoE version that doesn’t need 24GB+ of VRAM isn’t quite as good as the dense version, but it also doesn’t cost $200 a month with usage limitations and a company training their models on your data. If it’s anywhere near “good enough”, I can see this being a daily driver.

The business model should be that with economies of scale they could provide compute much cheaper than average consumer can buy to run locally.

That business model assumes that the huge cloud models will always maintain a gap worth paying for, compared to the local models. I'm just not convinced that the average consumer will need cloud models for summarizing their emails or the news of the day.

And for actual costs of their data centers, there literally aren't enough humans in the world where $20/month AI spending per person will help them break even. They'll need to sell big accounts (many businesses spending billions per year) in order to break even.