Qwen3.6 27B for running locally (16GB 4060 Ti) and I do often cheat and use Ollama Cloud to run gpt-oss:120B which is still surprisingly good but I've been trying to use GLM 5.2 lately just to try it out.

The trick is to pit these models against themselves as adversaries to spot hallucinations. That really improves the output.

Over time, though I've found that I've gotten better at using these models. Once you've got some solid .md files for context and you get used to the types of prompts that work best (with these models) I seem to be having less and less trouble over time. As in, I have to make fewer fixes or re-runs.

Note: You have to fix stuff constantly with the big models too. So it's not like any of them are perfect. You have to get used to using them effectively over time or you'll just end up with problems down the line.

Ugh. I'm sick of the reporting on data centers! There's only one (realistic) concern when it comes to data centers and it's this: Power consumption.

Everything else is either myths (e.g. water usage) or minor points that may or may not bother the locals (using up land that could've been a farm or a school or whatever).

Also, there's many kinds and sizes of data centers! Consider for a moment that nearly every school or modern business has a data center of some kind in their building(s). You have to put those servers somewhere!

People talk about data centers like the world just collectively decided that every single one that's ever built from now until the end of time will be as big as an airport and use up more power than a small city. That's not reality! Most data centers are just small portions of a larger business campus.

Yet now we're trying to regulate something so common and necessary that is going to pointlessly increase the costs of everything for everyone. Stop for a moment and think about it:

• Without data centers we wouldn't even have the Internet.
• Not all data centers are being used for AI stuff. AI accounts for a large chunk of BIG data center construction right now but that's not always going to be the case! Also, small data centers (in comparison) are the norm and have been being constructed for decades without anyone giving a damn or noticing their presence.
• It's cheaper for everyone to build a handful of really big data centers in a few strategic locations (i.e. where power is cheap and there's plenty of fiber) than a lot of big/medium data centers all over the place. This is also the most environmentally friendly way to do it.

If we want to actually do the right thing when it comes to data center regulations there's really only one thing to be concerned about: Power usage. It's also really easy to fix!

We just need a regulation that requires all data centers over a certain size to be powered via renewable energy. BAM! Problem solved.

It's just that no politicians have the guts to force such a thing and that too drives me nuts!

Solar is already the cheapest power generation option and there's plenty of space (in the US) to install more than enough solar panels to power all the data centers in the world (really, we have so much space!). Also, data centers can be anywhere! That AI video generation service could be out in the middle of the desert in Nevada and it wouldn't have an impact on the user experience in the slightest... Even if the user was in Australia (on the other side of the world)!

Big AI investors should be worried! I use AI every day at work (mostly for code-related things) and the open weights/free models are nearly as good as their ridiculously expensive services. Which are only getting more expensive over time!

Not only that, but the free stuff is improving at a faster rate than the models Big AI is putting out. The only thing keeping Big AI alive right now is coasting. Businesses haven't figured out that they can self-host the free models with their own (or rented) hardware for a fraction of the cost.

As soon as these businesses figure out that you can configure Copilot in Visual Studio Code to use FREE models that are 90% as good as Big AI, it's over for Big AI.

There's also dozens of great tools to run ChatGPT-like interfaces using local models and they let you switch which models you want to use on-the-fly. Also, we've got ZeroClaw now which works better with free/tiny models for most of the stuff you'd want to use it for.

Aside: Big AI also has a Big Problem with security that goes away if you use local AI models: You never have to worry about leaking credentials or proprietary/confidential information to the cloud/Big AI.

The chances this fixes it are... Remote.

While I sort of agree with you that that is his angle, service jobs that aren't easily automatable will continue to climb in the "job stability and pay rate" category until the world figures out how to deal with Baumol's Cost Disease:

https://en.wikipedia.org/wiki/Baumol_effect

Just in time for the Steam Machine and Steam Frame!

Right, Valve? Right?

...any day now...

Price fixing is not cool.

Fortipatch your Fortishit already!

Superman was in too many anti-drug public service ads.

People who undertook this hobby weren't quite sure where the pigeon would be when the pictures were taken. They just had to wing it.

Don't ask questions when you know the answer is fowl.