apitman

I think I misunderstood the original problem

The ESP32-P4 is a beast

Would a HDMI dummy plug work?

Have you tried hooking an LLM up to them to see if it can figure out what's wrong?

There are a lot more devices being released built on esp32s lately. e-readers are another example. I suspect with current hardware prices we might see a significant rise in consumer compute being limited to these types of devices, and a corresponding push to make software efficient enough to run on them. Could be interesting.

I think updating js without review is also against policy.

I am fairly certain this is against Apple's policies. In principle you're not supposed to change the functionality of your app without review.

I personally would love them

Makes me a little sad that it's been almost exactly 6 years since I made this comment: https://news.ycombinator.com/item?id=24475946

In that time I started a domain registrar, a free OpenID Connect login service (https://lastlogin.net/), created the de facto list of tunneling tools (https://github.com/anderspitman/awesome-tunneling), and built a couple of my own tunneling tools (SirTunnel and boringproxy), all to try and make self hosting easier.

I'm still trying to solve this problem. Some progress has been made (mostly by other projects), but we're still not there.

FWIW, I used to think everyone owning a domain was the way forward. I now suspect otherwise. Domains are too expensive (especially most TLDs outside .com/.net/.org) and the costs of accidentally forgetting to renew are catastrophic. Domains just aren't a consumer product.

I think you only need a domain if you want a permanent public web presence, and I don't think most people need that. Most people just need a better/faster/private google drive, google photos, google docs, etc, and for that it's ok if your domain changes once in a while.

So I think free subdomains set up without vendor lockin are probably the ticket.

If anyone is interested in collaborating in this space, definitely reach out.

> Self hosting implies having hardware and software knowledge to set it up

This is unfortunately true, which is why I think we need a different term for turn-key self hosting. I like indiehosting myself.

Indie hosting = owning a car so you can use it Self hosting = building a kit car

> I feel sad about the artists who suffer due to AI-slop competition

If it's able to generate that is competitive with artists, is it still slop?

It's interesting to see the definitions of terms like "slop" and "vibe coding" evolve in real time.

You could try renting a GPU or VPS

I've found agents very effective at both idebtifying root causes from vague descriptions, and explaining both the problem and the fix to me.

It's not a panacea though. Sometimes it comes up with bad causes and/or solutions.

This comment captures the current AI zeitgeist about as well as anything I've seen.

The way I look at these types of difficult debugging problems has completely changed.

If it's a system I want to understand more deeply and I have the time, I try to force myself to do it by hand. If not, the agent does it.

The trap is that it's easy to convince myself that I never have the time.

Has nobody from any of the companies hosting open weights models released detailed information on how much it really costs?

I'm not sure how useful a personal assistant would be if it doesn't have internet access, and at that point you're vulnerable to the lethal trifecta so I'm not sure local versus cloud model is the most important distinction to index on

It's Netflix before the unbundling happened

That's pretty sweet. Thanks

I'm sometimes tempted to use these sorts of proxies, but I always worry that the hacks they have to use to wrap the upstream tools and APIs is going to mess with my downstream harness/agent.

What do they do with old ones? Their hardware physically can't run other models right?

You can turn on various levels of some of many of the models in the UI

Seems like that equation really should weight total params and active params differently

Very interesting. I was not expecting anything close to this.

Currently software is king when it comes to inference performance though.

You should be getting way more than that on a 6000 pro even today. I'm getting 40tok/s on a pair of 3060s. You can ask a SOTA model to optimize your setup for you.

Aren't things like KV size inherent to the model?

I wonder how this would stack up against 4x RTX 3060, assuming you have the physical room for them.

Interesting. Why don't the unsloth guides (https://unsloth.ai/docs/models/qwen3.8) mention this? Do they already include the fixes in their GGUFs?

Check out OpenCode Go as well. They give some Kimi K3, Qwen3.8 Max, and GLM5.2 (probably 5.3 soon?) usage which may cover your needs for $10/mo