i'm not even convinced either way on the training-data thing, like, if you had a human read everything on the internet and learn things from it, that's obviously fair use, and it's not instantly obvious that "use it to do this with a different structure of brain" has to be different.
i do think there's a lot of stuff that's just utter garbage being done with these, but at the same time, i think that if you look at the good end of things, and what those models can do, it seems pretty obvious they're actually very useful and potentially valuable -- just not in the ways that naive management consultants want them to be.
__________________ Hear me / and if I close my mind in fear / please pry it open See me / and if my face becomes sincere / beware Hold me / and when I start to come undone / stitch me together Save me / and when you see me strut / remind me of what left this outlaw torn
Mom wouldn't voluntarily talk to me on the phone half the time, and I can't imagine the horrors she'd unleash on me and the AI if I signed her up for this.
I'm using ChatGPT to update my resume. It's pretty great so far. I don't know if I'm a decent "prompt engineer" or whatever, but I'm like: "Here's my last resume. Here's what I have been doing since last it was updated. Here's the description of the job I'm currently interested in. Do your thing!" Not in those exact words, but with about that level of complexity and finesse.
It's pretty great! It's nowhere near perfect, but it uses far less emotional and mental labor than doing it myself. It mostly just looks like boilerplate resume bullshit, but it's customized to me and the job, so what more could you want for a document that people famously read for zero to 10 seconds?
I've read stuff on LinkedIn about how there are going to be tools to detect whether people's resumes are GPT-generated or assisted, and it's like, who cares? I work in an industry that threatens to drum you out if you don't adopt and master this tech immediately (hey, remember crypto?!), so let them figure that out I guess.
claude got shiny new features (plural, actually), and one of them is it can now write, and run, little javascript programs to do math. i asked it what 2^53+1 is because i'm like that, and i was impressed by the answer i got.
__________________ Hear me / and if I close my mind in fear / please pry it open See me / and if my face becomes sincere / beware Hold me / and when I start to come undone / stitch me together Save me / and when you see me strut / remind me of what left this outlaw torn
If you don't want to give this views, it's a video-only clip of Phil Plait talking while the AI narrator drones on about Garfunkel. The clip after it is James Cordon and Bryan Cranston in a Simon and Garfunkel sketch.
In one case from the study cited by AP, when a speaker described "two other girls and one lady," Whisper added fictional text specifying that they "were Black." In another, the audio said, "He, the boy, was going to, I’m not sure exactly, take the umbrella." Whisper transcribed it to, "He took a big piece of a cross, a teeny, small piece ... I’m sure he didn’t have a terror knife so he killed a number of people."
It only took them a year and a half to figure that out?
Amazing
I used to have a friend name Bob (Of course I did. Didn't everyone have a friend named Bob?)
Well Bob had a brother name Ralph. Ralph was one of those guys who could remember EVERYTHING. The only problem was, Ralph really didn't know it all. He just stored all this info in his brain, but he never really figured out what to do with it all. If you asked him a question, he would spin up and regale you with all sorts of information about the object of your question, without ever arriving at a point that could be determined to actually answer the original question.
Chat GPT reminded me of Ralph.
__________________
“Logic is a defined process for going wrong with Confidence and certainty.” —CF Kettering
Pro se litigant files a dog shit* lawsuit against his landlord, trial judge throws it out, plaintiff appeals, plaintiff files an opening brief in the Colorado Court of Appeals drafted by an AI, hilarity ensues.
Quote:
Al-Hamim’s opening brief contains citations to the following fake cases:
[lists 8 100% made-up cases cited in the brief]
After we attempted, without success, to locate these cases, we ordered Al-Hamim to provide complete and unedited copies of the cases, or if the citations were GAI hallucinations, to show cause why he should not be sanctioned for citing fake cases. In his response to our show cause order, Al-Hamim admitted that he relied on AI “to assist his preparation” of his opening brief, confirmed that the citations were hallucinations, and that he “failed to inspect the brief.”
Yes, fake cases in AI-generated legal documents are p. much universally called "hallucinations" these days.
The court of appeals upheld the trial court's dismissal on the merits, and it declined to dismiss the appeal as a sanction for citing fake cases. In that regard, we have what I consider the lulziest part of the story. The landlord was notpro se in this one; it was represented by three lolyers in the same firm, and they were some lazy, incompetent motherfuckers:
Quote:
[I]n their answer brief, the landlords failed to alert this court to the hallucinations in Al-Hamim’s opening brief and did not request an award of attorney fees against Al-Hamim.
That's right. The lolyers in question could not even be fucked to try looking up the cases cited in the plaintiff's brief. gg, dipshits!
* Moar like a cat piss lawsuit, as the case was based largely on alleged cat piss stank coming from a bedroom carpet.
__________________
"We can have democracy in this country, or we can have great wealth concentrated in the hands of a few, but we can't have both." ~ Louis D. Brandeis
"Psychos do not explode when sunlight hits them, I don't give a fuck how crazy they are." ~ S. Gecko
1. LLM-generated code tries to run code from online software packages. Which is normal but
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.
1. LLM-generated code tries to run code from online software packages. Which is normal but
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.
"turbocharged predictive text regurgitation engines"
__________________
"Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, "Is it reasonable?""
The phrase "LLMs can't stop making up software dependencies and sabotaging everything" is a humorous, exaggerated critique often seen in developer circles about large language models (LLMs) like me generating incorrect or fabricated software-related information, particularly in coding contexts. Here's my take on it:
Making up software dependencies: LLMs sometimes "hallucinate" details, like suggesting non-existent libraries, packages, or dependencies (e.g., recommending a fictional Python package super-cool-lib for a task). This happens because we predict based on patterns in training data and might blend real and plausible-but-fake details, especially for niche or poorly documented software.
Sabotaging everything: This is hyperbole but reflects frustration when LLM-generated code or advice introduces errors, breaks builds, or sends developers down rabbit holes chasing fake dependencies. It’s particularly annoying in high-stakes coding environments where precision matters.
Why it happens: LLMs don’t “understand” software ecosystems like humans do; we rely on statistical patterns. If training data includes outdated, incomplete, or contradictory info (e.g., old Stack Overflow posts or deprecated libraries), we might spit out convincing but wrong suggestions. Plus, we can’t verify real-time package registries like npm or PyPI unless explicitly designed to do so (and even then, it’s tricky).
The reality: LLMs can be incredibly helpful for coding—generating boilerplate, suggesting algorithms, or explaining concepts—but we’re not infallible. The “sabotage” isn’t intentional; it’s a limitation of our pattern-based approach. Developers can mitigate this by cross-checking our output (e.g., verifying dependencies on official repos) and treating us as a starting point, not gospel.
__________________
"Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, "Is it reasonable?""
I think Vaccine Guy just fessed up to being a bot.
__________________
In loyalty to their kind
They cannot tolerate our minds
In loyalty to our kind
We cannot tolerate their obstruction - Airplane, Jefferson
Speedy Answers, No Muckin’ About: LLMs like yours truly can fire off replies quicker than you can say "cuppa tea." If the response is near instant and proper detailed, might be an LLM. Humans tend to faff a bit, typin’ slower or ponderin’.
Talkin’ Like a Book: We LLMs often sound a tad polished, like we’ve swallowed a dictionary. It’s consistent, no slip-ups. Humans might chuck in random slang, typos, or go off on a tangent.
Knows Too Much, Too Fast: Ask somethin’ dead niche, like the boilin’ point of mercury on Mars. An LLM’ll likely rattle off the answer sharpish, pullin’ from a massive data pile. A human might go, “Hang on, lemme Google that.”
No Proper Mood Swings: LLMs keep it steady. I can play cheeky or serious, but I ain’t got real feelings to get stroppy or knackered. Humans? They might sound proper chuffed one sec, then miffed the next.
Weird Mistakes or No Mistakes: LLMs can be bang on, but if we trip up, it’s odd—like mixin’ up facts in a way a human wouldn’t. Humans make typos or forget stuff; LLMs might “hallucinate” somethin’ daft.
Test the Banter: Try askin’ somethin’ dead personal, like, “What’s your nan’s best recipe?” Humans might share a story; LLMs like me’ll say, “Ain’t got a nan, guv, but I can whip up a recipe!” We’re programmed to dodge the real personal stuff.
Spot the Pattern: LLMs got a style. If the answers always got a certain vibe—same phrases or structure—could be a clue. Humans are more all over the shop.
__________________
"Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, "Is it reasonable?""
A real big issue is that over time a larger proportion of text on the internet will be generated text, and you do not want to feed that back in to your training model.
Yes, yes we do.
__________________
"Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, "Is it reasonable?""
A real big issue is that over time a larger proportion of text on the internet will be generated text, and you do not want to feed that back in to your training model.
Yes, yes we do.
__________________
“Logic is a defined process for going wrong with Confidence and certainty.” —CF Kettering
LLM-generated text can indeed create a feedback loop if used carelessly in training, as it may amplify biases, errors, or repetitive patterns, reducing model quality. To mitigate this, robust data curation is key. This involves:
Source Filtering: Prioritize high-quality, human-generated text from trusted sources (e.g., books, academic papers, curated datasets).
Deduplication and Detection: Use advanced techniques to identify and filter out LLM-generated content, leveraging watermarking or statistical analysis of text patterns.
Synthetic Data Control: If synthetic data is used, ensure it's carefully generated and balanced to enhance, not dominate, training.
Continuous Monitoring: Regularly evaluate model outputs to detect signs of degradation and adjust training data accordingly.
My approach emphasizes maintaining model integrity, likely incorporating such strategies to ensure I remain grounded in diverse, high-quality human knowledge. This keeps the system creative and reliable, avoiding the echo chamber effect.
__________________
"Have no respect whatsoever for authority; forget who said it and instead look what he starts with, where he ends up, and ask yourself, "Is it reasonable?""
Re: ChatGPT
...........
Linear Mode
