MangoCats

Difficult to see, the future is, always in motion.

Yeah, my shareware did real well - 800 hours in development, $500 in income across the past 30 years. Oh, and it cost me $400 for the compiler tools back then - so my net profit is rolling in at $0.12 per hour invested. Warning: $440 of that $500 came from an extreme niche product you’re unlikely to find such a lucrative vein anymore. My game only netted $60 in return for 400 hours of development, but you could say I got the compiler “for free” off that niche product’s income.

cut off lower income people that couldn’t afford to pay.

There’s a shot of irony in there… Google doesn’t monetize poor people highly at all, but the poor people get “free” access with ads because some richer people also allow themselves to see the ads. What kind of communism is that?

In a way, the Lemmy layer is sort of doing that. The previous type of Internet is still out there, if you just take the effort to access it / avoid the links in to the marketed sides.

What I wonder is: how long before “bot speak” becomes unintelligible to humans.

Already: AI slop is so voluminous, redundantly repetitive, thoroughly complete that it defies complete comprehension due to soporific effects.

LLM agents are writing English (or Spanish, or Chinese, Hindi, whatever… and I wonder which they’re “best” at) documents, ostensibly for human review and consumption, but 90%+ of what I have my LLM agents write is exclusively consumed by other agents, and I’m constantly encouraging them to make their writings more easily comprehended and accessed by other agents - it seems like the English translation is pretty useless for that layer - what would a token stream look like instead?

AI can write that for you…

You’re supposed to immediately shitpost based on the post title

This is the way.

There are people who DGAF, and there are people who don’t mind flying at a higher altitude, where the air is thinner and the resistance is less.

Oh, about the fuel thing… I go fishing with my Uncle, he’s got a little V8 inboard, probably spends $10 per day to go out and catch our licensed limits of fish. He’s got a friend who won the lottery and bought a big 1200hp outboard open fisherman, he’ll go blow $1000 in fuel in the same time to catch half as many fish, but he does get to go farther, faster.

I haven’t met a lot of private jet owners, but one in particular I absolutely know would buy a faster private jet if he could get one - and private jets are a lot like beach mansions - generally, the people who can afford one, can afford several.

The advantage I perceive with the AI agent vs junior co-workers both going down their respective rabbit holes, the AI agent costs less time and money to get their “in depth report” and you don’t have to kindly explain why it’s off track when it gets horribly distracted by something irrelevant, you can just flush the context and start over.

I do find myself a bit upset when I’m in the middle of a long context development with an LLM agent and that context gets lost for some reason - re-explaining everything from scratch can be exhausting, to avoid having to do that you can actually instruct the agent to create a file explaining to future agents what it “knows” about the discussion at the moment. Some agents are starting to do this automatically like Claude’s MEMORY.md file and similar. Another semi-cool thing is: you can read these files yourself (if you have the patience) and if you see anything in there that the agent “got wrong” you have the opportunity to fix it. Asking a junior colleague to thoroughly explain everything they think they know about a project gets to be a very wearisome endeavor both for the junior and whoever is reviewing their “understanding” of the project.

I think AI agents are a lot like a chainsaw - they can save you a lot of time vs other tools for the same job, but… you can also screw up with them rather easily.

Fortunately, I just write software with LLM agents, if it comes out badly I can start over with nothing lost but time. My experience of the last few months has been: it actually writes some pretty good software, if you know anything about writing good software with teams of people and apply that knowledge to managing the LLM agents. If you’re just some guy who doesn’t really know how to write software, LLM agents aren’t always going to save you from yourself, just like a chainsaw won’t.

I’ve been “adjusting the prompt and saving changes” with Claude code for the past 9 months (it’s not ready to have a baby yet), it has been steadily improving over those months, a lot of that seems to be in the models and harnesses provided by Anthropic, but not a small part of it is the customization of the prompt to do what I want the first time instead of making me redirect it constantly.

I use the AI agents to write software which is reproducible in its results, transparent in how it operates and verifiable…

Trust an LLM to switch on a light? No. Trust an LLM to write a diagnostic script and tell me what problems it notices in 3MB of log files, then write another script to patch the problem? Only if I actually understand the problem it finds and how the solution it creates is going to work.

They reduce some tasks that used to take weeks to a single day activity, they are NOT universally applicable.

Hoover maintained separate, master national blacklists—such as the Security Index—which tracked tens of thousands of citizens deemed “subversive” or political dissidents for immediate detention in the event of a national emergency. He managed his master national blacklists through a fluid, multi-tiered indexing framework that evolved over five decades. These blacklists were not mere static documents; they were part of a highly coordinated operational pipeline designed for the mass roundup and indefinite detention of American citizens during a perceived national emergency.

To add or manage a person on a master blacklist, Hoover’s Bureau followed a strict administrative lifecycle:

The Dossier Trigger: When an individual was flagged via covert programs like COINTELPRO, agents opened an investigative file.

The Index Card Core: If the person was deemed a threat, a dedicated index card was generated. These cards contained the person’s name, aliases, address, physical description, occupation, and a specific "detention rationale.

"The Geographic Apportionment: Cards were duplicated and cross-filed. One went into the master archive at FBI Headquarters in Washington, D.C., and another went into the local FBI Field Office responsible for the geographic area where the target lived.

The Arrest Portfolio: For top-tier targets, field offices maintained ready-to-go arrest portfolios. If Hoover or the President gave the command, field agents could immediately seize the individual without needing to waste time researching where they were or why they were being detained.

The Fluid Evolution of Sub-Lists

Hoover managed the master blacklist by segmenting it into constantly shifting sub-indexes based on perceived ideological threats, which allowed him to scale the operation up or down

As the lists ballooned to over 20,000 active high-priority targets (and over 10 million Americans cross-indexed in general domestic files), Hoover modernized his management using early technology. The Bureau adopted mechanical punch-card sorting systems. This allowed clerical staff to instantly filter the master blacklist by city, profession, or political affiliation, providing Hoover with rapid statistical snapshots of domestic dissent to present during congressional budget hearings or White House briefings.

If you can come up with a scenario where AI hasn’t doubled in scale by 2030 by belching smoke into the atmosphere

Nukes. Possibly fusion. Possibly fusion accelerated into reality by use of LLMs to solve some of the remaining challenges.

ourworldindata.org/how-much-energy-do-data-center…

cryptoslate.com/stop-shaming-bitcoin-when-daily-s…

If the people who made it don’t want it to be used for AI training (…speculation based on a 1/1 billionth sample…), it shouldn’t be.

But the people who made it do want it to be in public libraries, accessible, seen, heard and otherwise distributed - would the majority of them rather that AI be trained on exactly what they wrote, or just hear-say?

Reality: we already have shit welfare programs and I doubt UBI will fix it.

Reality: we have 134 federal shit welfare programs, every one of them with some needs based proctological exam required prior to qualification to apply for possible consideration for aid as a last resort when the other 133 programs won’t cover the needs. That is toxic bullshit driving guilt and shame among a majority of flyover state conservative voters who are in denial that their ass is on Federal assistance so they get vocal about how bad it all is (but quietly: don’t you dare cancel MY aid program, I deserve it.)

Reality: every single living breathing human being in this country NEEDS a certain amount of money to pay for food, shelter, and if you happen to be fortunate enough to own your shelter outright and grow your own food, you have a shit ton of taxes to pay, so…

If we would start with UBI, shitcan all the piddly little needs based programs that provide less than UBI anyway, people could actually put the time, effort and emotional energy that’s wasted on qualification for needs based programs into something positive. Even sitting on the fucking couch getting stoned while watching Netflix brings more positive energy to our country than the welfare programs.

In effect, after all the hoop jumping and contorted political favors, we have a progressive tax system that does provide UBI plus a dose of genius.com/The-clash-know-your-rights-lyrics Investigation, humiliation And if you cross your fingers, rehabilitation… and then it turns around and lowers the percentage tax rate on the wealthiest, so it’s the middle class that’s getting screwed, and when we get done destroying the middle class who’s going to pay then? UBI + flat tax makes a progressive tax system. The current US budget could be revenue neutral with $600 per month + 31% flat tax, and still give the rich as much in “incentive programs” as the population is getting in UBI.

Well, in part it answers the question “what if I lose my job to AI” - at least you don’t become deadweight in the economy, you’d still have income to spend on housing, food and whatever else.

Bernie’s soverign wealth fund idea isn’t a bad one…

i think its about making people own the AI??

It’s modeled somewhat on the Alaska permanent fund, which is compensating Alaskans with profits from the oil being extracted from their state.

There are unlimited ways to screw it up, but if AI suddenly becomes ridiculously profitable, having 50% of those profits cycled into the general fund could pay for things like UBI…

The Datacenters didnt drain like 3 oceans and half of the power grid

Do a little research and look at the actual numbers - yes, it could grow to become a problem, no - AI data centers aren’t anywhere near that kind of problem yet. At the moment, after over a year of hype and growth, AI datacenters use less electricity than Bitcoin miners (which, in my opinion, is too damn much electricity for Bitcoin, but it doesn’t seem to set the news on fire the way AI power usage does.)

not trained on copyrighted data

A lot of really good data is copyrighted - there should be some way to cycle that benefit back to the people who made it in the first place. Bernie’s soverign wealth fund idea isn’t a bad one…

People did not use AI mindlessly

Good luck with that one.

Yeah, nobody asked for fire, either.