The Day I Almost Got Rich 😄

Screenshot of my mining dashboard during the session — hash rate swings, accepted shares, and difficulty stats all in one place 😄

Every miner has that moment staring at the dashboard, wondering: “Did I just get close to a block?” This is my little LBD (Looking Back Data in this case, although perhaps you're a little [LAPSE... brain dead] from reading it!) post on one specific session where I hit a Best Share of 145.5M. What makes this run even more fun is that it was completely solar powered — running off a GoLabs R150 power bank fed by a Renogy 100W panel plus a Volt Hero 20W panel wired in parallel for a combined 120W of solar input. Off‑grid mining at its finest 😄(* “free” electricity to run the thing, but it cost me $150+ of equipment to get there). And all of that powering a tiny Bitdsk N5.Rex ASIC miner — a little 9‑watt unit that somehow manages to push out around 168.8 Gh/s while sipping power like a phone charger.

The Snapshot

Three Local LLMs Walk Into a Pipeline: An LBD Deep‑Dive

Three local AI models get the same systems-engineering homework. The results are... educational.

When you feed the same complex systems-engineering prompt into three very different local LLMs, you learn a lot—not just about the models, but about how modern AI handles the difference between understanding an architecture and actually implementing one.

The prompt itself was non-trivial: build a fully asynchronous Python data-ingestion pipeline using asyncio, httpx, Pydantic v2, a custom token-bucket rate limiter, exponential-backoff retry logic, a bounded queue, and a worker pool that maintains thread-safe moving averages.

At first glance, there was a clear winner (at least according to Gemini's quick review). Then I looked more closely at the winner.

And that's where this little experiment got considerably more interesting.

Below, we break down how each model responded, what Gemini observed, what I found when I checked the generated code more closely, and what the results actually say about local LLM engineering.

The AI Warning-Label Challenge

Every now and then, my wife throws me a challenge that is equal parts mischievous and revealing. This time, she hit me with the infamous prompt:

“Describe me like you’re trying to warn someone about me before they meet me.”

Naturally, I accepted — and then escalated it. I ran the prompt through four different AI systems: Copilot, Gemini, Claude, and ChatGPT. What followed was… well, a mix of flattery, psychological profiling, technical audits, and mild character assassination. In other words: a normal Tuesday. (Except it's currently Saturday. I also had Copilot draft this blog post, including the quotes below from each AI. Because it's what I do. Sometimes.)

Below is the full set of responses, presented verbatim, each in its own block. If you’re new here, this is exactly the kind of thing my LBD experiments tend to produce — unexpected, slightly chaotic, and probably something I’ll regret publishing. (Oh, and apologies for the occasional stray moments of enhanced vocabulary in the AI replies - I don't talk that way, even to the AIs, but sometimes they go their own way. I'm just quoting exactly as they replied below.)

Dombot V2: When a Fictional World‑Domination AI Starts Rewriting How It Thinks

If you’ve been following the Dombot Experiment, you know the premise: build a fictional “world domination AI,” give it a multi‑phase master blueprint, and let it iterate on the instructions governing its own output.

With Dombot V2, we introduced the biggest architectural change yet: Dombot can now rewrite the prompts it uses to generate each phase.

Put simply:

V1 asked: “What should Dombot write?”
V2 asks: “What instructions should Dombot use to decide what to write next?”

That shift — from autonomous content generation to autonomous methodology generation — is the conceptual hook of the entire experiment (so far).