@jaredwhite one hell of a hell

@yomimono did you like AND share the post?

An observation on Sam Altman’s talks on AGI is how there was a focus on hypothetical harms from AI destroying humanity but a lack of acknowledgement let alone concerns about today’s real world harms ranging from impact to white collar jobs, to teen mental health and misinformation.

The era of tech acknowledging externalities of their products and staffing teams to address them now seems to have been a zero interest rate phenomenon.

I think the disappearance of these kinds of trades feeds the feeling that the modern world is terrible and strange strains of nostalgia that right wing politicians seem adept at exploiting and turning in to much uglier things... without ever bringing back any of the things from "the good old days" that were actually good.

Like shoes that you could love and have for 25 years.

I've heard people suggest that products that last slow economies. That isn't true. They change who gets to make money.

Of course, what they really mean is "we need power for the arms race". Whatever AI can do, we want to be the ones to scale it up the mostest.

That assumes that current scaling laws will hold 10 years from now, which is a big bet.

But even if they do, do we really want new emergent abilities as fast as we can possibly get them? Our society has barely adapted to the introduction of social media, let alone AI.

"McDonald’s tried an AI-driven system for its drive-thrus for three years before ending the effort. During those years, the system made mistakes like trying to add bacon to an ice cream order and giving one customer an order of 260 Chicken McNuggets."

@carnage4life

Can't wait for "DivorceGPT"

New from 404 Media: Republicans are trying to cram a ban on AI regulation into the Budget Reconciliation bill, which would stop states from regulating AI entirely for 10 years.

https://www.404media.co/republicans-try-to-cram-ban-on-ai-regulation-into-budget-reconciliation-bill/

We are oddly selective about the government services about which we will say, they operate at a loss.

Commuter rail, we are often told, operates at a loss. But we spend even more on streets and highways, and take in little direct revenue from them. Yet no one ever says they are operating at a loss.

@verge this sucks; I hope some day it will be made available as a standalone game

From my very casual experimentation, it seems like Claude may be better at vibe coding than at debugging or small tasks. Ask it how to write a single Pandas evaluation expression that has some complex conditional logic and it sends you up completely the wrong alley. But ask it to build an entire mapping web app using Leaflet from scratch and you'll have a thing that pretty much works in less than an hour.

Hundreds of thousands of Computers won't be able to upgrade to Windows 11, but that shouldn't make them eWaste.

Kudos to the @kde team for this amazing initiative!

https://endof10.org/

@Loraine @yianiris kind of a tangent, but content can be both privately owned and publicly available; that's the case for any public code repository on Github which doesn't have an open-source license. it defaults to "all rights reserved", not "public domain."

I think I've finally figured out what puzzles me so much about the Fediverse approach to moderation and safety. There seems to be a widespread focus on identifying and blocking bad *people* and the places where they gather, rather than toxic *content*. The problem with this is that the latter is much less of a game of whack-a-mole and can even be automated to some degree, whereas toxic people can always make new accounts and find new places to gather.

Americans confused by the Canadian election need to learn the two types of democracies:

1) a Parliamentary system
2) a Funkadelicary system

In a Parliament, Dr. Funkenstein summons the Mothership to defeat Sir Noze

In a Funkadelic, freeing your mind allows your maggot brain to create one nation

@carnage4life Seeing "AI trainer" in the list of jobs that companies are advertising in lieu of prompt engineer makes me wonder if the shift towards local LLM usage has something to do with it. Prompt tuning was always presented as an alternative to fine-tuning and now that the latter is easier even on relatively large models the "one-shot obviates fine-tuning, bigger is better" sales pitch for API-based LLM services may be less persuasive.

@carnage4life it's crazy how many outlets are reporting on the visit with the headline that Vance met the Pope, which is apparently untrue?

Microsoft researchers create super‑efficient AI that uses up to 96% less energy
Future AI might not need supercomputers thanks to models like BitNet b1.58 2B4T.
https://arstechnica.com/ai/2025/04/microsoft-researchers-create-super%e2%80%91efficient-ai-that-uses-up-to-96-less-energy/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social

Philip K. Dickens 😂