Remember 100 years ago when radiation was new and exciting and poorly understood, and companies were putting radiation into everything?
Thankfully we learned our lesson and will never make that mistake again.
Remote
Chris Ammerman
@cammerman@mstdn.social
mastodon 4.7.2Hobbit tendencies. Indoor enthusiast. Books and games. Tired but not sleepy.
Software architecture. System design. Tidy first. Slow is smooth and smooth is fast.
Humane management. Sustainable business.
Democracy. People before property. Freedom from religion. We're all in this together.
Mostly cis. Probably straight. He/They
574 Followers
248 Following
33 Posts
Joined November 04, 2022
Replying to
Here's one of my prophecies about LLMs that has come true:
Getting good software out of an LLM is like getting good software out of an intern with no prior professional expertise. Technically possible, but it takes a *lot* of time and attention from the mentor.
And the downside is: This investment of effort when applied with a human eventually produces a human that no longer needs it. But an LLM will always need you to micromanage it and it will never learn from you the way a human would.
Hey. Nations are made up. Citizenship is made up. Naturalization is made up. We have these things because this is the social contract we grew over hundreds of years.
The idea of an immigrant becoming a citizen is no less legitimate than the idea of someone being born a citizen. They are both and equally legal constructs that exist because things were worse without them.
If an immigrant can be denaturalized, then anyone the government wants to be rid of can be denaturalized.
Replying to
I am even more convinced now that anyone who uses these things in a mostly hands-off mode cannot avoid producing balls of mud that will be impossible to untangle--AT BEST.
I don't believe it's possible to prompt in such a way as to avoid this. Some questions just require reasoning capability to know that they don't need to be asked.
LLMs are incapable of sense-making. All they do is superficial "looks like it could make sense." But the sense isn't actually there except by pure coincidence.
Replying to
No matter how many prompt contexts you stack up and fire off in parallel, the machine cannot find truth, cannot do math, cannot know things, or reason.
It's Massively Multiplayer Online Autocomplete.
The fact that the capital and executive class thinks this is sufficient to replace most of the world's knowledge workers tells you all you need to know about how we should be dealing with them, and all of this.
Replying to
I really do think that you can't trust an LLM unless you think these things don't actually matter.
And there are plenty of people who think that. I've worked with lots of them. To our collective chagrin, many of them manage to ship product.... It's just not product that I would be proud of or want to put my name on. But that doesn't stop them. It works (for some very accommodating definition of "works") and that's good enough for them and the bosses.
But not for me.
Replying to
Look, if you tell an LLM it needs 95% confidence, it doesn't know what either "95%" nor "confidence" means. It knows people tend to respond to this kind of direction either by saying "I'm not sure enough because..." or "I'm super confident for these reasons." It has no ability to correctly choose which of those templates it will follow.
Flip a coin. You'll get a reasonable looking sentence back in one of those styles, with a random assortment of reasons that may or may not be rooted in fact.
Replying to
You cannot reliably get an LLM to produce good code and good architecture. You can get it to do so *sometimes.* But you can't really control when it does.
Other times it is just going to produce things that superficially appear like they could be good code or good architecture. They wear the trappings, but the veneer is thin.
The rough shape of something that looks like good, is just not good enough to ensure a workable, maintainable, supportable codebase.
And I think that still matters.
Replying to
@mcc@mastodon.social I imagine a curmudgeonly old man holed up somewhere, probably around the Basque regions of France, where his family fled when Napoleon seized power. The last of his family line. Still holding the Revolution in his heart. Reckoning the days and years according to the revolutionary calendar. He grabs his smartphone to post today's rallying cry via his Mastodon server (vivela@revolution.social.) And he curses as he is confronted--yet again--by the Gregorian date on the lock screen.
Replying to
I guess there is a very short way to summarize what I'm saying here...
Many people want to know "can you get an LLM to produce good code and good architecture?"
For some people the answer is, "It doesn't matter." This argument is not specific to AI and the debate has been ongoing for decades, so we won't resolve it here.
There are a lot of people out there trying to argue that the answer is "yes."
And I disagree...
Even with all of these new developments, programming as a career and programmers as a class are still extremely privileged.
It's really important to remember that. Everything that the capital class is trying to claw back from us has already been taken from others, and has never been granted to the vast majority of people.
The rest of the working class have always been fighting this fight. We are going to have to accept and respect that. We can no longer pretend it's not our fight too.
Boosted by @ferrix@mastodon.online
I don't know how we ended up in a universe where "Trump Derangement Syndrome" refers to people who don't like Trump, rather than to all the batshit fucking unhinged behavior of all the people that have twisted their souls into pretzels supporting or following him.
As I remain committed to doing everything I can with my own hands on deterministic tools, I look forward to someday being the smartest one in the room simply because I have not gotten any dumber over time.
https://futurism.com/artificial-intelligence/ai-boiling-frog-human-cognition-study
I recognize this is not the most constructive attitude, but I refuse to be shamed by my energy company for "using more energy than my neighbors" to keep my house at a temp I like while I'm working from home every day, when our state and county governments are looking for every opportunity to bring in new data centers.
Fuck you. Build some nuclear and/or get the data centers the fuck out of here.
Replying to
An executive or senior leader who wants to track AI usage as a metric loses my confidence immediately. At *best* it's making a lot of unfounded assumptions. At worst it's just straight up magical thinking. If they are lucky they might get a stock price bump from talking about AI usage, or laying people off because the AI usage is so high. But you can't prop up the stock price indefinitely on vibes and no outcomes.
General artificial intelligence is not required for the current AI regime to result in a human disaster. The careless abandon with which engineers are hooking up unproven experiments not only to each other, but to existing trusted systems, seems monumentally careless and dangerous. I think something like a runaway train incident is inevitable.
Sigh. Companies continue to not understand software development in general, and software quality in specific.
https://www.theverge.com/transportation/956316/ford-quality-jd-power-ranking-ai-automated-mistakes
There is no new phenomenon here. AI takes a supporting role in this failure, but it's just another variable in the context of this mismanagement, not a root cause or primary factor.
Replying to
Regulate VC and PE.
Outlaw leveraged buyouts.
Regulate private currencies as securities.
Enforce monopoly and trust regulations.
Reverse Citizens United.
Prosecute insider trading and conflicts of interest in elected and appointed officials.
...For a start.
Replying to
@futurebird I think everything like this really just means "they're not one of us," and that's pretty fundamental to fascism.
Someone called "pastor" saying Trump understands the Bible better than the Pope is not actually a statement about Trump. It's just fundamentalists hating Catholics like they always have, and using that to fluff Trump, because he continues to give them progress toward their ultimate goal of theocracy in the US. Which is why "christofascist" is not an exaggeration, it's literally what they are.
Replying to
People think "good at arguing" means "intelligent." But "good at arguing" is like "good at handiwork" or "good with cars" or "good at cooking." It is a skill and a muscle. It *is* a kind of intelligence. It is a mental capability. It is admirable, and worthy of envy. But it isn't inherently transferrable. It isn't general-purpose genius, no matter how much they might believe it is.
It would be nice to be able to see why something appeared in my Mastodon feed. Like, what am I following that the post is part of.
Not that it would have stopped the unobscured penis drawing hashtag spam from popping up at the top of my feed, but at least I could consider unfollowing whatever brought that post to my feed.
Replying to
I say that as someone who I believe only very narrowly dodged that contrarian self-brainwashing. All the seeds were there. All the conditions were right for me to move from smug religious conservatism to smug contrarian atheism.
I don't know how I avoided it. I think I got lucky.
Replying to
Contrarianism is the refuge of people who are smart and mentally agile enough to be good at the mechanics of argument, but who are too privileged and lazy to bother building actual expertise.
Replying to
Frustrations today with #godot #GameDev in C#:
I've been using Chickensoft tools to help lay a foundation for working in VSCode. It's mostly been very helpful.
Except.
1) VSCode's test panel cannot see tests written using Chickensoft.GoDotTest. Results are console-only.
2) When debugging Godot code in Visual Studio, the debugger does not detach when the Godot process terminates, or when the tests finish.
I think #1 is just a weakness of the GoDotTest. Root cause of #2 could be anywhere.