Workshop notes

Four blue screens, one pattern

My brother's PC kept crashing — with a different error code every time. That was the clue: wandering codes are the signature of failing hardware. How an AI helped prove a known silicon defect.

My brother's high-end PC — just out of the retailer's warranty — had been crashing for days, and eventually stopped booting into Windows altogether. The photographed blue screens showed four different error codes. To the AI I showed them to, that wasn't noise; it was the finding: “Four different stop codes within a few days — the pattern itself is the diagnosis.” A single broken driver crashes in the same place every time. Codes wandering across core Windows components are the signature of failing hardware.

The first ranking of suspects — memory on top, the processor near the bottom — survived exactly one reply. Then the CPU model came up, and the AI turned the case around: “That changes the diagnosis fundamentally. This isn't just any CPU — that's Raptor Lake Refresh.” This processor generation has a documented series defect: the chip requests too much voltage at idle and ages irreversibly, until it miscalculates under everyday load. The AI searched the web as we went — and found, along the way, that Intel had extended the warranty for exactly this defect by two years. “Just out of warranty” had only ever been true for the retailer.

Prove first, replace second

Suspicion isn't proof. The AI wrote a small diagnostic program that collects everything with evidentiary weight: crash logs, unexpected reboots (fifteen in two days), memory dumps — and the microcode level, the microprogram the BIOS feeds the processor at every boot and through which Intel defuses the defect. That field came back empty, a bug in the AI's own script. So it decoded the four raw bytes by hand — and found the real knot: the machine was running a microcode level from before all the fixes. The mitigation had never arrived.

That reordered everything: update the BIOS first, judge afterwards. The update turned into an adventure of its own — wrong USB port, a temperamental 256 GB stick, a status LED that glowed solid for twenty minutes while nothing happened. In that moment the AI's main contribution was being the calm voice: “Hands off. Don't cut power, don't pull the stick, don't press any buttons.”

After the successful update, on default settings and without memory overclocking, the proof arrived on its own: a fresh crash, 21 seconds after boot — on a microcode newer than every official fix level. The AI's verdict: “That is the irreversible degradation. Microcode doesn't heal it. The CPU needs to be replaced.”

The outcome — and what stayed with the humans

A local shop installed a replacement processor shortly after; the machine has run without a single crash since, and the memory test we ran later passed clean — the counter-check that finally cleared the RAM. The evidence package prepared for Intel's warranty process was, in the end, deliberately never filed: for my brother, the effort wasn't worth it. The value lay elsewhere — a solid diagnosis instead of days of guessing, and a computer that was back to work quickly.

Every decision stayed with a human: which steps to take, what to buy, what a formal process is worth. And oversight was needed, too. The AI started with the wrong prime suspect, caught two bugs in its own script — and stubbornly kept referring to an “8 GB stick” that never existed; it was the old 256 GB one with a shrunken partition. My correction, verbatim: “There IS no 8 GB stick … you pest ;)”

The same series defect had, as it happens, hit my own laptop a few weeks earlier — there the processor is soldered in, and the story played out very differently: Instability is poison.

/compact — the essentials when context is scarce:

A high-end PC crashed with four different error codes — wandering codes are the signature of failing hardware, not of one driver. The AI recognised the documented defect of that processor generation, discovered the vendor's two-year warranty extension, and built a diagnostic program that delivered the proof: a crash 21 seconds after boot despite up-to-date microcode. Replacement CPU installed, stable since, memory test clean; the formal warranty claim was deliberately skipped. Along the way the AI erred too (wrong first suspect, an invented USB stick) — the value came from diagnosis plus pushback, not blind trust.

An unhandled error has occurred. Reload 🗙

Rejoining the server…

Rejoin failed… trying again in seconds.

Failed to rejoin.
Please retry or reload the page.

The session has been paused by the server.

Failed to resume the session.
Please retry or reload the page.