Wes Deviers

Technologist, nerd, refined rural citizan. I only post in good faith.

[Wes Deviers @wesdeviers]

@BREGOH @system_monarch Relax... The target audience is people who previously didn't know load balancers work at multiple layers. It'll be okay.

[Wes Deviers @wesdeviers]

@WWolf287 @LgclLibertarian @LPNational Sorry, how old are patents again?

[Wes Deviers @wesdeviers]

@J_Von_Random @StabbyTheCrab @esrtweet Nope; as I read it, you were responding to the "datacenters can be replaced by local hardware" thread. If I misread that, though, I'm sorry.

I remember similar predictions from when public cloud exploded. "Oh it's just temporary, soon we'll all run K8S clusters at home.."

[Wes Deviers @wesdeviers]

@J_Von_Random @StabbyTheCrab @esrtweet Nothing about that scales, or passes the "husband" or "wife" test:

Will my non-nerd [husband/wife] tolerate a world in which my homelab runs his/her "hey chat?" and is down all the time?

[Wes Deviers @wesdeviers]

Solid take. In the meantime, senior engineers are going to have to wheelbarrow a lot of tar to stay employed.

[Bruce F. Webster @bfwebster]

Would be interested in your critique on this message I sent back to my son-in-law, a senior QA engineer whose company is drinking deeply of the AI-Will-Fix-Everything koolaid.
—-
SIL: Can I ask your honest opinion on AI? Where is this going to land?

Me: Here’s where I think things are headed.

There are a number of firms and individuals who are learning how to properly leverage AI and are doing amazing things with it.

Everyone else who is using it are setting themselves up for serious problems, especially since a lot of them won’t recognize those problems until damage has been done.

I think general public use of AI is going to run into the “unreliable narrator” pitfall: trust what the AI says and does when the information may be biased, misleading, or just plain wrong. This is not just the hallucination problem; it’s also biased or wrong information getting baked into the LLMs (eg, how much LLM training has been done based on Reddit threads).

SIL: Some of my managers seem to ignore things like Skill Atrophy. I had the view that those in senior positions now are ok because they have the experience to help understand how to fix the issues with AI Coding. However, they seem to think those issues are just a matter of training AI to do better. But, skill atrophy is real, developers are writing about it and I know I've read some papers about studies into it. Things like Critical Thinking are decreasing because people rely on AI to do it for them.

So I wonder if AI will really overcome these issues and become just as good if not better than humans at this. Are we going to see a takeover of engineering by AI?

Me: No, because I think those efforts will lead to massive technical debt and serious maintenance problems. But I fear things will get ugly before then.

Look, as it is roughly 2/3rds of software project are late, over budget, underperform, or fail altogether. As an industry, we’re just really bad at large software projects. AI will solve some of those problems at first, or at least appear to, but because of all the human reasons we’re bad at software, I think a lot of firms will end up in a AI legacy tar pit that will be hard to climb out of.

[Wes Deviers @wesdeviers]

@howard You're assuming quality matters. I argue that's a false assumption.

[Wes Deviers @wesdeviers]

@D_E_Harris_Best @AP I had a really long answer to this, but then I realized I didn't want to give unimaginative people ideas.

I'll just stick with "you're not wrong, but your assertion is incomplete."

[Wes Deviers @wesdeviers]

You unfortunately just proved an opposite case: you can buy all of that stuff from any mail order place w/ no background checks practically anywhere in the US. And yet.... and yet.... nobody is out there car jacking with a replica '58 New Army. Despite the fact that it's a perfectly fine weapon to carjack somebody with. Why.... why do you suppose that is?

[Wes Deviers @wesdeviers]

@GetonTrightside @FenixAmmunition The average datacenter is smaller than the average Walmart.

[Wes Deviers @wesdeviers]

@DP_LTS @joquinn12345678 @GetonTrightside @FenixAmmunition Neither of those things are true. But specifically you don't need low latency for surveillance. You need throughput.

[Wes Deviers @wesdeviers]

It's tedious in the same way writing fiction or reviewing bad movies might become. Writing enterprise software professionally is like reviewing Suburban Sasquatch because you need to pay rent with Patreon.

When you write code for fun, you figure out how to fix the tedium. LLM dev is also fun for hobby use.

[Wes Deviers @wesdeviers]

@ZappoMan @VicVijayakumar This is true for now, but it's a limited time window. Those things will stop mattering soon, just like DevOps and public cloud made infrastructure not matter. Not that it's better, just that in the end the people writing the checks don't care.

[Wes Deviers @wesdeviers]

@engineers_feed I do 40+48, then subtract 3.

[Wes Deviers @wesdeviers]

@SpencrGreenberg Language is the core of power - there's a reason people argue so much over it.

You do great work, btw. Glad I ran across the youtube channel.

[Spencer Greenberg 🔍 @SpencrGreenberg]

@PaulaSeeksTruth It’s a good question but it’s hard to answer in the abstract since there are many types of unreasonable.

[Wes Deviers @wesdeviers]

I think the fault in your logic is that you assume owners like PE care about long term quality. But they don't, at least not generally. Flip the companies. Nobody gives a crap about technical debt or tech due diligence. Make garbage with AI, sell the company before the architectural debt comes due. That's already part of the playbook. AI accelerates the decline in quality... But only (unemployed) engineers care about quality.

[Danger Casey @CaseySoftware]

I worked on a smaller scope of this problem ~20 years ago at the Library of Congress

Our mission was to digitize everything the Library held BUT make sure it was recoverable for someone 100 or 1000 years ago

To emphasize the point, my boss would bring two items to meetings: a handwritten captain's journal from a ship in the late 1700s and an 8" floppy. He'd pull out both and say "one of these is 200 years old, the other is 20. We can only read one of them"

What we quickly decided is that EVERY digital format would become replaced and potentially unreadable over time

Therefore, for the digitization steps, we would document the entire process - equipment models, settings, etc - and capture the uncompressed, raw output (tiff, wav, etc). The physical media was destined for an N-filled vault in Culpepper, VA

The raw masters were kept unedited for preservation but then downsampled into smaller, more mobile formats (jpg, mp3) for presentation

The goal was that every X years, archivists could pull the masters and update them to a new masters (likely keeping the old) and then generate then-modern compressed formats

Thinking about "how does someone read this file format in 100 years?" was a *really* cool thought experiment to put into practice

And btw, yes sometimes finding equipment to play the physical media was an adventure unto itself. The wax cylinders were the most novel and using images to "play" a record were the most novel but Edison's first motion pictures were something else entirely..

With that "something else" including nitroglycerin..

[Wes Deviers @wesdeviers]

You have to backtrack further into the past than that. The circumstances matter - did anybody force either of them to go to the bar? With each other? To keep drinking? To walk back to the room?

If we hold people accountable for drunk driving, then we should hold people accountable for other actions taken while drunk. If you are going to say, "She was too drunk to know what she was doing" - I think that is a perfectly reasonable thing to say. But that's not how our society works. In EVERY OTHER CASE you are accountable for what you do while in altered states.

[Wes Deviers @wesdeviers]

@Ashley78186050 @leftlateral There's not actually much of America where that kind of crime happens very much. And nicer cars tend to be in nicer (low crime) neighborhoods. The thing about crime is that it tends to be hyper local.