trey-jones

Well, yes, it is a dumb TV - that still seems normal to me. I don't like computers in most of my things. I like them where I can see them.

I have a server running in my basement with 2 nvidia Tesla GPUs that I bought for ~$70 each on ebay. (Unfortunately the RAM cost an arm and a leg). I built it for inference, but I also set it up to run sunshine server and now I have my Steam library on the TV through a Raspberry Pi 3. Even better, it was a fun and interesting project, maybe not as straightforward as you think. Unfortunately it's entirely possible that it costs .24/hr just for the electricity.

Now, this feels fine with <10ms input latency over wifi. Even in 2026 I'm skeptical about sending my input to Atlanta and back.

I always thought this was a hot take but I agree!

Yes - the older I get, the more obvious it is that the mind of a child (and young adult) is better suited to learning, reacts faster, and is more creative than mine.

The attention-deficit that is so common among especially (but not only) young people today means that grabbing their interest is of paramount importance to their learning. Just repeating that kids are bad at reading stuffy texts. But if you can put them onto something that they find interesting, they can definitely excel. I've got tales of high school students who are terrible students, but give them a CAD and 3D printer and they will solve any problem you have.

So - in other news, education is broken. How do we fix it?

This is exactly true in my experience. I find myself trusting the LLM better than I ever trusted a development team to deliver accurate work. Just like you would managing a team of junior developers, you have to keep an eye on the output, but I think the agent is more transparent about that than your team would be. And it works faster, and you don't have to deal with emotional meltdown or active sabotage if you criticize in the wrong way.

The trick for the guy controlling the agent is to know about the gotchas that one would normally encounter when doing the work. Asking the right questions and mentioning the right things to pay extra attention to can make all the difference.

I think that's why it will be much more difficult for people to learn how to make software now. Those of us who have been doing it for years and years have experience that is valuable in steering the LLM. It will be hard for people new to software to get that same experience. But then things are changing so quickly right now, everything I just wrote my be irrelevant in a week.

I won't read anything on twitter as it currently stands, because the product sucks. The "fixed" functionality provided by xCancel is what makes it worth it for me to read. So, if you want me as a customer, you need to provide similar functionality.

Not all customers are as particular as I am, therefore a shitty product continues to thrive, and I can go do other things.

Regardless of the legality of the service:

I always use xcancel instead of X. I don't have a twitter account and I don't want to sign in. I just occasionally want to read what some people had to say about something. Even this is usually a mistake. It's probably better taken as entertainment.

My point is: companies stop making your product suck, and people won't have a reason to try and fix it for you.

I've been using 5.3 since they initially announced it and my gut feel is that it's not as good as 5.2 for agentic tasks. I'm still using it - I don't think it's bad. I'm just not convinced it's better.

This is exactly how I feel about my job now. I'm the new manager. But honestly claude is better than any team of developers I've ever had working under me. It does exactly what I ask it to, in 10 minutes instead of 3 weeks. And apparently we can't afford a development team now anyway...

I'm not working at massive scale. I've not built any public-facing projects from scratch. But I have done years-worth of cleanup and improvement on projects that I'm very familiar with and have been in my backlog. It's not 100% success, but it's better than I could have done on my own, in the last few months anyway.

I used claude with GLM and it's easy to set up, just hard to find the documentation. No headaches really, unless you want to use it against multiple different APIs.

Definitely feels like a lot more than 2x to me. Any time it feels like the agent is taking a long time I have to look in the mirror and say to myself "What could you have done in 5 minutes?"

When I found Tesla P100s on ebay for $75 I thought I was getting a cheap server, but 1/3 of the total cost was RAM. Sorry, first time I've built a computer since the shortage.

I don't know why we're even talking about Qwen3.6 for writing code when qwen3-coder exists. My experience is there's no contest. I'm using 30b with 96k context on a dedicated server.

Qwen3.6 was the first model I ran locally that seemed smart, but qwen3-coder:30b is way, way more responsive and more accurate for writing code according to my tests, including human-eval. If you can run one than you can almost certainly run the other. If you haven't tried qwen3-coder I would definitely recommend it. It feels positively snappy compared to every other local model I've tried. All you need is 32G VRAM and some heat dissipation.

You don't have to have a monopoly to be monopolistic.