Claude Code is revolutionizing digital archaeology. Enterprise better dig it

OPINION The UK's National Museum of Computing at Bletchley Park (TNMoC) is a natural center of digital archaeology. It has restored, recreated, and documented landmark machines, from the wartime Colossus computer through to British mainframes of the '60s and '70s. Its collection from the dawn of the microprocessor age also …

[just4this]

Seems unnecessary.

The origins and inner workings of all computers originate from Ancient Aliens.

[deadlockvictim]

Re: Seems unnecessary.

I prefer to think of the first maths professor in Queen's College, Cork: George Boole.

https://en.wikipedia.org/wiki/George_Boole

[Pete 2]

Decompiling Windows

> In the case of badly documented, poorly understood, and generally mysterious digital technology, we have the proof of concept that LLMs cut the mustard.

I look forward to the day (hopefully soon) when somebody gives Claude a copy of Windows to chew on, and asks it the question: Why would anybody do this?

Although if Claude either responded with I can't think of a single good reason or simply crawls into a metaphorical corner and starts crying, I would not be surprised.

[GL1zdA]

This is the way

Last weekend I've patched a ROM of a GeForce4 MX 440 to work via DVI with a modern 4K screen. It would fail to negotiate a reasonable resolution resulting in a black screen. I fed Claude Code to bad ROM, a ROM known to boot well, EDIDs of serveral displays, gave access to Ghidra via MPC. It already new the nouveau repository and envytools, which helped immensly to pinpoint what is happening. It managed to code a test harness to test the ROM with the EDIDs, produced some small diagnostic programs to run on the actual machine to observe what is happening and produced a 7 byte patch (3 procedures were patched). And it works.

Someone else on vogons enabled the line-in of an Audigy 2 ZS under DOS.

All these things bring retro computing to a whole new level.

[Bebu sa Ware]

"Here's our server map, document it and tell me what to do"

Luxury !

You had a map.

I joke not. A map implies the insanity might be represented by a planar graph and not need some ten dimensional string theory.

[that one in the corner]

Digital Archaeology == pattern matching & assembling at the overlaps

Machine Learning has been used, successfully, for a good long while doing similar tasks in a variety of other fields. Patching together sections of DNA into a complete genome by pattern-matching sequences of bases to determine where they overlap and can then joined - whilst allowing for the individual differences (from copy errors and, well, samples from multiple individuals). Similar approach would[1] work for growth rings, where individuals differ due to local effects but can be combined to show accurate larger range effects.

This is the bread and butter of ML: putting in the machine cycles to perform work that logically *could* be done manually, but where the cost/benefits make no sense[2].

Digging through documents, source code, random notes somebody has left etc and assembling them via the overlaps and subtle patterns is another good example[3]. Add in testing and review by humans and the ta-da.

BUT don't go around touting this as "LLMs are super for every use case, they generate so much original material after really understanding our situation".

Here, they *want* old stuff dug up and regurgitated. The current-day equivalent is to spit out the most-repeated StackOverflow answer about the barely-out-of-nappies shiny tool du jour.

Here, they *want* to see somebody else's original, thoroughly copyrighted, materials given a quick polish and presented on a silver platter. But does your company's new project want to risk using IBM's source code[4] with all the serial numbers filed off? What one pattern-matcher can discover for you, another can discover for the IBM lawyers once your project is released (wittingly or not[4]).

But even here they fall into the trap of seeing slice'n'splice as comprehension of the material. Sigh.

[1] "would" because I have no references for this; yet. Anybody?

[2] in the modern world; the ladies who pored over astrophotography plates looking for subtle differences achieved great science but are now replaced by ML

[3] although the resource costs not borne directly by these specific researcher may reduce the "goodness".

[4] coming to the Adelphi, the hilarious rewrite of the classic stage play, we present "No data leaks, please, we're British".

[Doctor Syntax]

"Forensic science owes a great deal to archaeology, such as excavation methodology, taphonomy – the study of what happens to organic remains after death – osteology, and the analysis of mass graves and conflict sites."

As someone who has been involved in both not only is that rue but the overlap is much larger. The big similarity, which applies here too is "You have to make do with what you have". The quality control lab in the glass works, for instance, can take as big a sample as they want, the forensic scientist has just tiny fragments and nothing more. It's all about pushing analytical techniques. There have been - maybe still are - university departments combining the two.

[BoHu]

"datacenters are the only known way of completely destroying water" (32:30)

That Rob Smallshire experience ("full story" TFA link) is interesting indeed, using AI (so-called) with 40 years of software dev and teaching experience (the YT with the 'big AI skeptic' comment at 33:32). It seems more sensible than newbies mindlessly 'vibe-coding' themselves some useless app, imho.

ISTM he's putting some agentic harness girdle infrastructure together, Fantasm (https://github.com/acornaeology/fantasm/tree/master), with DASMOS tracing disassembler (20:00), some Control Flow Graphing stuff (CFG) to de-linearize code (21:25), and eventually an annotation algorithm (23:10) that starts with CFG leaves and proceeds in reverse breadth-first order, calling the LLM to output human-readable comments (NLP), rasing the level of abstraction of the disassembly (in a sense). Perty cool.

The 41:30 Q&A of why he didn't use the NSA's Ghidra Software-Reverse-Engineering (SRE) framework and/or the Capstone (ultimate disasm engine) disassembly framework was neat too. My question would have been: given all the useful and efficient procedural stuff that he's put together for this, could he envision replacing the trillion+ parameter mammoth-bottomed LLM with some much slimmer and humane NLP system for generating sensible code annotations? Maybe that could be an upcoming improvement to the system, and might even help resolve some 'context indigestion' issues (16:50) ... ;)

[drankinatty]

I call B.ewildering S.tatement

"... Claude Code ... interpreting circuit diagrams ..."

Anyone that has ever struggled reading hardware datasheets and circuit diagrams probably has a good idea of the mental gymnastics involved. While coding models like Claude can churn through source code quite well "interpreting circuit diagrams" isn't something that falls within tokenizing, transforming and inference. Now I'm not a AI user, other than dabbling to see what they could do, but unless there has been wild leaps forward in what they entail, I think that statement needs a bit of further qualification. I'd also be interested in a reply from anyone that has first-hand experience with a model interpreting circuit diagrams with anything other than rudimentary IC chips?

Just what is the article talking about there?

[Erik Beall]

Re: I call B.ewildering S.tatement

I was stunned when I finally tried using Claude code with kicad only a week ago. It is better than the average mid level EE primarily because of the total discipline it can muster. Context is critical as always and using it well is truly exhausting, just like any other good usage of AI where you are the loop (and costs very little when avoiding loop slop). It can't do trace layout (it can actually do pretty well for very simple circuits but it's variable, don't bother with even mildly complex) but it can create board outlined and place things in suggested good places that I often have to move around a bit, sometimes a lot, but usually it's within a cm of where it ends up. It catches subtle errors only a good, experienced EE would routinely catch. But you have to tell it to explain what it finds and read ALL of it, or it might miss the context. Exhausting but much wow.

[Dan 55]

"impeccably structured and commented series of source code files, much better than... before"

Or unfortunately you often find it produces wordy, verbose comments which are just tiresome to read and hide what you really need to know because you can't see the wood for the trees.

e.g. at 2:28 - every line gets a comment, no matter how unnecessary.