I’d like my Pip-Boy in matte black please.

I’ve written my own genealogy app as well, it has reinvigorated my efforts. I have a multi-level data model (evidentiary below, curated above) that has helped me piece together people across a multitude of disparate sources. Claude has been my sidekick, and it’s amazing- especially for the more toilsome tasks like extracting people’s details from newspaper articles.

Hey, just out of curiosity, what exactly does this mean?

    > (evidentiary below, curated above)

Specifically what is the above / below part?

Like if you said "curated below, evidentiary above", what should I understand has changed?

I have split up the data model into a handful of layers in order to separate concerns. The evidentiary layer are the raw bits - the media, sources, clippings, photos, etc. From there I derive "personas", which are the (incomplete and possibly inaccurate) representations of people from those media bits. Above that is the curation layer, where the curator takes those personas and compiles persons from them.

So if there are a bunch of articles that talk about someone's birthday but a couple give conflicting data, that is all present and recorded in the evidentiary layer, and it is the responsibility of the curator to decide which is the actual birthday for that person.

edit: the "below/above" terminology is a dependency relationship. The facts undergird the conclusions drawn from them. To flip that relationship would be to "retcon" facts to fit the conclusions a curator has already drawn, which would be a mistake.

Their infrastructure is buckling under increased demand from AI generated code.

Imagine the software development lifecycle is a river blocked by several dams- necessary processes that also slow the water’s volume and speed. Writing the code was a very big dam upstream, and successive dams didn’t need to handle a lot of water. With that dam busted wide open, the torrent of code is inundating downstream processes. GitHub happens to host many of them.

Earlier this year the GitHub COO shared a post on Twitter with explosive numbers they were facing. I cannot seem to find it at the moment. (Edit: shared in other replies to this thread.)

I mean, that’s what they say but it doesn’t pass the sniff test. Which other major service struggles so much with basic scaling? They’re not serving streaming video, they’re not a real-time system. Any problems scaling should have been fixed years ago. Especially with all of Microsoft’s resources behind them.

> they’re not a real-time system

I would argue they are. You can't cache a git repository and send people what it looked like a minute ago, because someone might have pushed 2 seconds ago.

I think I’ve seen that movie.

> it just reads as more evidence that LLMs are better at replacing leadership than anybody else.

I believe this counts as having “more confidence in AI’s advice than in that of colleagues”

leadership as a category functions on low-to-minimal subject matter expertise. you don't want c-suite merging code into production, they should be handling the business interests and direction based on their training and general, high-level, trickled-up understanding of how teams are cohering on strategy

that generalized, trickled-up understanding is also something that's much more easily replaced by an LLM. it was easy for me to spin up a program governance skill paired with a deterministic set of tests in a few weeks. it crawls team channels, project trackers, etc, identifying date drifts, updates, and missing milestones. add an analytical layer of why and how it all fits together in the roadmap and you've functionally obsoleted a lot of the non-technical bureaucrats whose only purpose is to tell someone what other people are doing - from Director to C-Suite, from C-Suite to shareholders

of course you want to keep the people around with the technical depth to construct a roadmap but when was the last time senior leadership in your org actually did any of the substantive work on roadmaps? it's all team leads with the leadership layer serving as the arbitrary rubber stamp on the plans someone else created

when it comes to knowing that there's a dependency on the component you're altering the data contract on, the LLM is useless unless there is pre-existing context or unless you command the LLM to burn your pile of tokens on it tracing every bit of logic in your codebase. unless you want to do nothing but this level of discovery all of the time, you'll need an engineer with lived experience of working in the codebase. here this averaged understanding is useless - you need context, you need granular expertise, all things an LLM fundamentally lacks for your specific codebase and your specific requirements

In a 30M+ LOC repo we blew out Qwen3.6’s context thinking about a code review. This was via direct call to model. Do harnesses facilitate better context management, or is there something else to accommodate its smaller context window?

What was it reviewing? Was it just a "take a look at this pile of code over here"? Because I think asking any model to review 30M lines of code is a stretch...

in general with automated code reviews I've found it fruitful to tell the model what specifically to look for and where. I usually don't get over 10% of the claude opus context window for code reviews, but it of course depends on how your code is structured, how much the agent has to explore etc.