Dean Sheppard

Engineer Republican 2A believer 1A defender

[Dean Sheppard @SheppaDean]

The property has water and electricity. What about a septic system? That’s the most important part that most people don’t realize. A new septic system requires engineering plans, soil testing, and, most important, township approval. Buy this property. Then wait for the township to say ‘nope’ because the water table is too high or a river is too close. And even if everything goes smoothly, it will cost at least $30k to install. So yes, $150k seems cheap but perhaps there’s a reason why it’s cheap.

[Dean Sheppard @SheppaDean]

@YonovFilip Codex can collaborate with other LLMs without any issues. In my setup, I configured Codex to use a local Qwen3.6 endpoint for validation/second opinion. Sometimes Qwen finds flaws in Codex’s approach, so Codex implements Qwen’s proposals instead.

[Dean Sheppard @SheppaDean]

That has nothing to do with grounding. See those thick traces? A lot of current flows through them. Bad or dirty contacts increase resistance, which generates heat and can burn the board. Inspect the mating pins for oxidation or corrosion. Also check that the board is seated firmly - the pins must have a good grip. If anything is loose, it will heat up and eventually burn.

[Dean Sheppard @SheppaDean]

@codevsdev This one was very popular back then.

[Dean Sheppard @SheppaDean]

@farlander762 @IanCutress I think they use a low boiling point Fluorinert, like FC-72, which has a boiling point of 56°C. That’s why you see the bubbles.

[Dean Sheppard @SheppaDean]

@cjzafir Are you serious? Using a local LLM to search for files - that’s the use case? Instead of find . -name "filename", you’re suggesting prompting a model to run the exact same command?

[Dean Sheppard @SheppaDean]

@RespectfulMemes

[Dean Sheppard @SheppaDean]

@georgecursor The first thing I did was vibe-code a service that uses RGB fans as indicators for llama.cpp health and GPU load. Blinking red = model crashed or not loaded, and a blue-to-red gradient indicates GPU load, with red representing 100% utilization.

[Dean Sheppard @SheppaDean]

@scottastronaut Very impressive! How well do the copper surfaces hold up against atmospheric oxidation?

[Dean Sheppard @SheppaDean]

@realmaximusfps @loyalmoses You’re right! With all the AI breakthroughs, it’s easy to forget how much work it would take to create a realistic animation.

[Dean Sheppard @SheppaDean]

@cartierfamilyZ Increasing revenue by 12,000%? Are they talking about startup growth here?

[Dean Sheppard @SheppaDean]

@loyalmoses Brass flying is definitely VFX (there are some inconsistent shapes) but slide recoil and animation look pretty realistic.

[Dean Sheppard @SheppaDean]

@samhogan No, we did not!

[Dean Sheppard @SheppaDean]

You don’t do much coding, do you? I’m a developer, so for me an LLM’s coding capabilities come first. I’ve tested most of the available models on real coding tasks - for example, giving them a non-obvious bug in a codebase and asking them to fix it.
One example was a lag issue in a Swift app when displaying a modal sheet. Qwen3.6-35B identified the problem and fixed it. Gemma-4-31B found the issue but couldn’t fix it. All the smaller or older models went completely off-topic.
So yes, for tasks like writing emails, summarizing text or booking appointments, a 9B model is probably sufficient. Not for any serious coding work.

[Dean Sheppard @SheppaDean]

In reality, Gemma-4 is absolutely not designed for agentic work that requires a long context. It’s a chat model with broad general knowledge and decent inference speed. All it can really do is talk - and it talks a lot. It’s good for short, one-shot tasks, but once the context grows, it tends to fall into hallucinations and looping behavior.
So, I’m sorry, but unless you can demonstrate an actual application where this model excels, I’ll consider this post clickbait.

[Dean Sheppard @SheppaDean]

@Bertrand_Charp

[Dean Sheppard @SheppaDean]

@RatsTesla Don’t windshields use a thin Low-E metal-oxide coating? If so, it can reflect radio waves, which might not be ideal for the link.

[Dean Sheppard @SheppaDean]

@ArchSxlt @LostMemeArchive It’s probably not even the hardware that’s running Doom. While some modern 32-bit microcontrollers can run it, I don’t think that’s the case here. This has to be just a display interface connected to a computer via USB.