SparkyMcUnicorn

6 cameras, 10.54ms. All other metrics are 0-3.5%

Same as any other open source framework or library.

Calling that "lock in" is a stretch, but you're free to write everything from scratch if that's the way you roll.

I'm running frigate on an SFF computer I got off eBay for $100 with an i7-8700. It averages around 14 watts. OpenVino for object detection and intel-qsv hardware acceleration preset.

What lock in?

I use their AI SDK, but never touch vercel servers. It's just a unified interface.

> My experience is that without at least one rooster the hens won't lay well

This is just not true. Hens will lay just as many eggs with or without a rooster. And a rooster doesn't do a whole lot to fend off predators, from my experience.

Same. With the recommended settings, it got it right. I regenerated a bunch of times, and it did suggest Cathy once or twice.

R1 70b also got it right just as many times for me.

1. You can get a 50% discount via batching.

2. Give a few Sonnet or 4o input/output examples to haiku, 4o-mini, or any other smaller model. Giving good examples to smaller models can bring the output quality closer to (or on par with) the better model.

You're describing functionality that's built into Aider. You might want to try it out.

Aider also has a copy/paste mode to use web ui interfaces/subscriptions instead apis.

I definitely use and update my CONVENTIONS.md files and started adding a second specification file for new projects. This + architect + "can your suggestion be improved, or is there a better way?" has gotten me pretty far.

An additional information panel shows up on the right hand side when you're logged in.

They're using it via fireworks.ai, which is the 685B model.

https://fireworks.ai/models/fireworks/deepseek-r1

Kagi is using fireworks.ai according to the docs, which is the 685B model.

Kagi and Aider benchmarks definitely put R1 in the lead, but I have yet to try them out.

I went this route, but eventually went back. HA automations became more powerful/flexible, and node-red became hard to manage, not easily updatable on mobile, and a bigger point of failure than HA.

Home Assistant has been rock solid with zero downtime for over a year now, and that last downtime was due to a hardware failure.

While I was looking for how to swap out the backend to self-hosted, I noticed that analytics is hardcoded to true. Something to be aware of.

Found this the other day on GitHub and have been meaning to try it out. Guess I'll be setting up the backend and modifying the hardcoded config in the extension.

On the other hand, I guess with this it's possible to overfit a decent model to my org for $450.

Agreed on Kagi ultimate.

The agents (optional with a toggle) hooked up to the LLMs are fairly decent. They'll search, grab YouTube transcripts, read online dev documentation, etc.

Crypto investing is where the points really start to fly, mass clicking buy below $10k selling above $20k. Game finished pretty quick after that.

Edit: Just beat it in 19 minutes. Feel like you can't finish it much quicker than that without cheating.

Copilot does have vision support, but needs the extension for it.

https://marketplace.visualstudio.com/items?itemName=ms-vscod...

To each their own, and everyone's experience seems to vary, but I have a hard time picturing people using Claude/ChatGPT web UIs for any serious developmen. It seems like so much time would he wasted recreating good context, copy/pasting, etc.

We have tools like Aider (which has copy/paste mode if you don't have API access for some reason), Cline, CoPilot edit mode, and more. Things like having a conventions file and exposing the dependencies list and easy additional of files into context seem essential to me in order to make LLMs productive, and I always spend more time steering results when easy consistent context isn't at my fingertips.

AWS Graviton is ARM.

My experience is that having a team with mixed platforms has helped reduce deployment woes, with the rare platform-specific bugs getting worked out beforehand.

I often forget the name of things, sometimes even the big ones. GitHub search is one of the primary ways I rediscover them. "reverse-engineer API" returns mitmproxy2swagger as the third result, and this is how I found it last time I needed it.

It is a bit frustrating when a project on GitHub doesn't have good tags or searchable keywords, making it harder to find.

Indeed, that does not appear to exist. NuPhy Air75 HE is probably closest, but it's not split.

https://keeb-finder.com/keyboards?ms_layoutSize=75%25%2COthe...

If you wanted to try again, take a look at the Nuphy Air75 v2. Maybe with the Aloe or Cowberry switches.

I feel like you can always find a mechanical that you like, no matter how particular you are. It's just whether or not you want to go down the rabbit hole.

What if the contractor used an LLM?

I've used the logitech g305 for at least 6 years now, and thankfully that one lets you write a profile to the chip on the mouse.

On AirBnB just check the "Display total before taxes" box on the search results page.

Not including fees in the nightly rate makes sense as some are fixed rate, and having the option to see the cost for the total stay (including fees) solves the problem.

If it's a static site, you could move it to Cloudflare Pages to reduce the cost to zero.

This doesn't include all the benchmarks.

The one that really stands out is GroundUI-1K, where it beats the competition by 46%.

Nova Pro looks like it could be a SOTA-comparable model at a lower price point.

> Codegen tools today are still not great: The lack of context and not using LSP really burns down the quality of the generated code

Have you tried Aider?

They've done some discovery on this subject, and it's currently using tree-sitter.

This is a short and fascinating peek into that world: https://youtu.be/qr26jxPIDm0

It doesn't look this cut and dry.

M4 Max 14 core has a single option of 36GB.

M4 Max 16 core lets you go up to 128GB.

So you can actually get more ram with the Pro than the base level Max.