Comment Re: What's the motivation? (Score 1) 153
I don't see the basis of Funny, especially since there was a Japanese version of Slashdot that apparently had Unicode a long time ago. Lost code problem? Or was it actually a Shift-JIS nightmare?
Comment Re:Probably [lost marbles] by an AI agent (Score 1) 61
That was almost exactly my reaction to the story, but I think you should have gone for funny with it. I'm also not sure you should have called it an "agent", however. So I have supporting anecdotes to share from my AI-supported "programming" experiences...
My early experiments were mostly with ChatGPT and DeepSeek. My website was getting sick and the PERL/CGI was no longer allowed to run, so one of the many upgrade paths I explored involved moving functions, mostly statistical stuff, from PERL to JavaScript. This is when I started encountering the lost marbles problems. The first few iterations would work surprisingly well, but then it would start losing its marbles and various features would disappear, seemingly at random. Not sure I could figure out when this was going on since time seems so distorted these years, but I feel like it was around two years back.
More recently the server died completely. (Tripod's parent company Lycos was supposedly quite valuable a long time ago, though not nearly as valuable as the AI companies are supposed to be these days--but that's a fresh bubble waiting to burst.) So I wound up using the quasi-website aspect of GitHub to host my quasi-website. (Not quite trivial to modify the old JavaScript utilities for the new URLs.) I also decided to take another swing at the bigger problems, this time using Claude. Color my surprised or even amazed? Much more productive this time around. My "work" pattern this time involves short sessions, basically discussing features and data structures, followed by a minute or two of file generation by Claude, a couple of minutes of file installation, and then some testing. Pretty quickly matched and went beyond the existing PERL code, including apparently fixing a regex problem that had eluded me for a long time.
At that point I started worrying about Claude losing its marbles. The AI is quite willing to discuss the problem in terms of tokens, but it refused to give any hard limits and apparently has no way to assess if a "session" is close to reaching any of them. However it definitely described behaviors that sounded like losing marbles and was unable to suggest any good ways to detect such problems. And I think that is probably what happened in this story. Someone was updating code using an AI and at some point it passed its limits and started losing features. Who knows what else has gone missing?
Claude does have some meta-features for managing tokens, including compression, but it was not too helpful about assessing the risks. Instead it suggested starting a fresh session and prepared an interesting "transition" document that is supposed to describe the current state of the new system... But the threats of lost marbles remain and the threats sound quite similar to what seems to have happened in Outlook in this story... I feel like Claude's threats are only implicit because it won't clarify what they are or how to detect them...
(Just about finished with Microsoft Secrets about their software development processes a long time ago. Testing problems were prevalent and never really solved...)
Submission + - Europe: The World's Fastest-warming Continent (barrons.com)
fjo3 writes: The latest heatwave sweeping across Europe is a stark reminder that it is the world's fastest-warming continent, stretching into an Arctic that is heating at an even greater pace.
Britain, France, Italy and Spain have issued red alerts and health warnings for much of their territory this week as the region endures its second heat episode since May.
Britain, France, Italy and Spain have issued red alerts and health warnings for much of their territory this week as the region endures its second heat episode since May.
Comment Re:Raises hand (Score 2) 61
"I think Microsoft just aren't testing legacy Outlook properly anymore
CI/CD for the win.
Comment Re:Just accel the move from Blue to Red states (Score 4, Informative) 105
There are no high paying jobs in data centers, just destruction of quality of life for locals. Perfect for red states, they are accustomed to being shit on, they vote for it.
Comment Re:capitalism is for sellouts (Score 1) 105
"The worst people have the most money because they have no compunctions about harming others."
Yes, that's why capitalism must be regulated, precisely the opposite of what Republicans have pursued since the 80s.
Comment Re:24/7 round the clock surveillance is abuse (Score 1) 94
That would have been an interesting angle, but I don't see 24/7 as the crux of the problem. The police-state/authoritarian personality is not crucially dependent on surveillance. If that were the case, then East Germany should still be going strong.
I can actually recall a stop-and-frisk scenario that convinced me the cops can find SOMETHING to make an issue of if they search carefully enough. Asking for a friend who feels lucky the police settled for a hundred bucks?
My own feelings are mixed. I'm a big believer in the truth and I don't have sufficiently negative words to capture my true feelings about liars. However I also think there are cases of "You can't handle the truth" and some of these cases might even involve police officers.
Comment But it's only a joke Subject (Score 1) 157
Still spanned about a quarter of the discussion...
Comment ACs can't smell (Score 1) 149
That's why they stink?
Oh, wait. I meant "spell", but they and their "you can't blame me if you don't know who I am" ideas do stink, too.
I should include some flavor of the old joke about mud wrestling with pigs, but that would take effort and the propagation only spanned about 1/6 of the discussion (by ye olde scrollbar metric), so such effort isn't justified.
But a joke related to the story? Can AIs solve the AC slop crisis? Or a joke about prison for ACs, coming real soon if'n AC actually lives in the wrong place.
Just joking. It's already arrived in a couple of places. I just don't want to name them because I might get put on a (yet another?) list.
Submission + - Helion says the 1st fusion power plant is coming soon. A cofounder isn't so sure (scientificamerican.com)
tedlistens writes: The startup backed by Sam Altman recently raised $465 million, tripling it's valuation as it races to build what it says will be the world's first fusion power plant, supplying Microsoft with carbon-free electricity in 2028.
But one of its founders—the plasma scientist whose research inspired its reactor design—has serious doubts.
But one of its founders—the plasma scientist whose research inspired its reactor design—has serious doubts.
Submission + - Bypass the polirical parties, add a new feedback to Congress (taxnvote.org)
SysEngineer writes: How would you change the US Federal budget? TaxNVote.org allows you to adjust 9 or 1000 categories of the next federal budget. The default form shows nine top-level categories (Defense, VA, Education, Health, Infrastructure, Science, Environment, DHS, Other); expand any line and you can allocate down to individual federal accounts — NASA, the National Park Service, specific research agencies, anything Congress votes on. Takes about five minutes at the top level, longer if you want the detail.
Tax N Vote (TNV) is a proposal to add a new feedback channel to the federal budget process. At tax filing each year, every taxpayer optionally submits a Tax Dollar — one person, one allocation. The IRS anonymizes submissions; the Census Bureau processes and stores them (where you can verify your own); the CBO aggregates one-person-one-vote between April 16 and May 1 and publishes "The People's Budget." A third reference point alongside the two party platforms — measurable, granular, and updated annually. Congress is not bound by it; what changes is that deviations from constituent preferences become documented, attributable, and electorally citable. The argument is system-dynamics, not partisan: changing the color of the players doesn't change the system. A simulation of the mechanism shows convergence toward whatever the People's Budget turns out to be, in both ideological directions tested. There will be a talk on the model at ISDC 2026 in Delft.
The Government-side processing of Tax Dollar documents is written in Rust — memory safety and predictable performance for government data handling. The browser-side allocation engine is a Rust WASM module inside a Vue frontend, so the math you see in the app is the same math the aggregator uses. Processing is divided across agencies that already exist; marginal cost to the government is less than renaming the Department of War.
Open source end to end. The Tax Dollar format is open, the reference implementation is at github.com/greenpdx/TaxNVote26, and anyone can build their own client, audit the aggregator, or publish pre-filled template budgets that citizens adopt with one click. Go build a budget: TaxNVote.org.
Tax N Vote (TNV) is a proposal to add a new feedback channel to the federal budget process. At tax filing each year, every taxpayer optionally submits a Tax Dollar — one person, one allocation. The IRS anonymizes submissions; the Census Bureau processes and stores them (where you can verify your own); the CBO aggregates one-person-one-vote between April 16 and May 1 and publishes "The People's Budget." A third reference point alongside the two party platforms — measurable, granular, and updated annually. Congress is not bound by it; what changes is that deviations from constituent preferences become documented, attributable, and electorally citable. The argument is system-dynamics, not partisan: changing the color of the players doesn't change the system. A simulation of the mechanism shows convergence toward whatever the People's Budget turns out to be, in both ideological directions tested. There will be a talk on the model at ISDC 2026 in Delft.
The Government-side processing of Tax Dollar documents is written in Rust — memory safety and predictable performance for government data handling. The browser-side allocation engine is a Rust WASM module inside a Vue frontend, so the math you see in the app is the same math the aggregator uses. Processing is divided across agencies that already exist; marginal cost to the government is less than renaming the Department of War.
Open source end to end. The Tax Dollar format is open, the reference implementation is at github.com/greenpdx/TaxNVote26, and anyone can build their own client, audit the aggregator, or publish pre-filled template budgets that citizens adopt with one click. Go build a budget: TaxNVote.org.
Comment Re:Yeah..... (Score 3, Interesting) 56
You don't know what the alleged patents are, or whether they are granted rather than just filed. If "everyone already does this", where this is what is claimed in the patent, then there will be documentation. If there is documentation, the patent will not be granted. It's not magic.
Comment Re:Isn't Robert X. Cringely a pseudonym? (Score 1) 56
"I ran headlong into what we now call hallucinations in 1996..."
Seems highly unlikely. "what we now call hallucinations" is a 21st century phenomenon, a decade later. "what we now call hallucinations" is an LLM failure mode, LLMs first appeared two decades later.
I wrote software with bugs back in the 80s, perhaps I have your hallucination claim beat by a decade, given that the term can mean anything.
Comment not AI then (Score 1) 56
"The reason 2Brains doesn't lie and the reason it's cheap are the same reason. It looks the fact up instead of guessing it
Then it is constrained by what it can look up. A search engine with a natural language interface, not AI.
The reason humans do not lie (except when they do) is because they have values, not because the world is a multiple choice test with a cheat sheet. Humans can show their work, this doesn't even do work, it isn't a solution to anything and a proper solution obviates the need. Why employ "reasoning" when all the answers already exist? Because they don't. But hey, we can see why this guy got out of the business, and with the big money got back in. Same shit, different day.
Also...
"It is the whole ballgame for enterprise AI."
No, it is not. The "hallucination problem" is a symptom of a grotesque failure of architecture, but the ballgame for enterprise AI is predicated on a lack of such failure. Fixing the "hallucination problem" means you're in the ballgame, not that you've won. AI companies aren't interested in fixing it, though, they're interested in a race to grab the cash. Enterprises need to wise up, these tools aren't being developed to do a good job but to make billionaires richer.
Comment Re:Industrial scale [percolation?] (Score 1) 74
A coffee snob? Just the human to ask in lieu of an AI (which will just tell me whatever it thinks I want to here).
I've been wondering whatever happened to percolated coffee. I'm guessing it tastes bad, but I didn't start drinking coffee until decades after I last saw a percolator.
