zach

I work in industrial mechanical and process construction as an ML engineer. We build your favorite companies data centers.

[zach @0xperez]

@ClaudeDevs This is great but this would never fly for companies who are in compliance driven environments EVER.

[zach @0xperez]

Man the flat API pricing for Claude Enterprise is absolute BS

[zach @0xperez]

@SenTomCotton God you’re a legit an idiot.

[zach @0xperez]

@media_saboteur @MartinShkreli @HarryStebbings Lmfao what the fuck are you smoking and can I have some

[zach @0xperez]

@media_saboteur @MartinShkreli @HarryStebbings To me that just sounds like a poorly structured HR department. Ideally, unless you need to talk to them, you never know who they are, and there are no meetings. They are there for confidentiality and to manage employee benefits.

[zach @0xperez]

@bubbleboi this is beautiful fan fiction man

[zach @0xperez]

The Number They Don’t Measure

https://t.co/eaBEfzsob1

[zach @0xperez]

@iamgingertrash Consciousness and goal-oriented processing co-emerge from embodied control loops. Agency is homeostasis extended through time. LLMs trace the trajectories but lack a body, budget, and stakes, so competence without care.

[zach @0xperez]

@FallowWing @Jonathan_Blow @FrameworkPuter Yes, but it’s still technically optional metadata. You just won’t get the auto-request for admin rights. I’m not defending Windows, but he definitely overgeneralized, he made it sound like XML is a required hard dependency for all Windows executables, which just isn’t the case.

[zach @0xperez]

@Jonathan_Blow @FrameworkPuter that’s just the manifest. It’s existed for 20 years and is optional. Windows doesn’t “need XML in executables” for basic features.

[zach @0xperez]

@attentionmech the desert fathers are the only ones to figure how to venture into the unconscious from a conscious practice IMO. sit down, focus on deep breath, and then release the gates. let all thoughts through and label each thought as they come (e.g., “story”, “doubt”, etc.)

[zach @0xperez]

@snats_xyz @Dorialexander the cooling systems that are being installed in these new data centers are closed loop cooling systems. they are filled once. if you wanna riot about water waste go to your nearest power plant, they all have cooling towers that use open loop systems (constant consumption).

[zach @0xperez]

@SmokeAwayyy You do realize that they don’t actually build the data centers right, they contract the construction part out to a GC that hires subcontractors. They just ensure the people constructing it follow their specs (often done by an engineering firm, outside the client).

[zach @0xperez]

@VictorTaelin the technical quality is incredible. keep up the work man. maybe the message wasn’t quite heard from the AI labs perspective? maybe the issue is the pitch itself because from the outside looking in, you have the tech.

[zach @0xperez]

@JamesParadigms @theo There is no “fee” because they don’t exist. You can’t pay for something that isn’t there. I had to interview over 50 people to replace a data engineering position, mostly new grads. The multi-generational Americans were so bad I just could not hire them. And im American.

[zach @0xperez]

@benhylak that’s because most of them don’t care about software engineering. This generation was marketed internships for $100k+/year just by showing up with the degree. I blame the big tech companies for pushing this so heavily during the 2010s.

[zach @0xperez]

@zanehengsperger I’ve never seen one that low before. You can make more money working the night shift at Whataburger.

[zach @0xperez]

@CSCawley @btbytes @zanehengsperger This is the dream, we actually manufacture our own duct from raw sheet metal. The issue is we build any mechanical system in all commercial + industrial settings except oil. It would be impossible to manufacture every aspect of what we do.

[zach @0xperez]

@btbytes @CSCawley @zanehengsperger This is what I’ve always heard, but then you’ll often hear that RAG limitations are quickly met, which seems to be rooted in data quality issues. Manufacturing and Construction would massively benefit from unsupervised data labeling models that understand their tribal knowledge.

[zach @0xperez]

@TrueAIHound Can’t you know this but still use them to regurgitate immediately available general information that would otherwise take longer to collect and digest through some google deep dive? To me they just are text corpus alchemy machines; kind of up to the user to determine usefulness.