this exists it is called thinking
September 20, 2025 at 12:15 PM
this exists it is called thinking
They're gonna feed this into an LLM and then ask the LLM who to fire based on the answers. Answer accordingly!
NEW: The email asking fed workers what they got done last week is out. It asks "What did you do last week?" and says "Please reply to this email with approximately. 5 bullets of what you accomplished" (but no classified info). Deadline is 11:59 p.m. Monday
www.washingtonpost.com/nation/2025/...
www.washingtonpost.com/nation/2025/...
Feds must answer email on what they did last week — or lose jobs, Musk says
Musk says on X that feds must explain what they did last week — or resign
www.washingtonpost.com
February 22, 2025 at 9:51 PM
They're gonna feed this into an LLM and then ask the LLM who to fire based on the answers. Answer accordingly!
My understanding is that the evidence shows LLM-based assessment gives better 'marks' to LLM-written things (essays, grants etc). So by choosing to use LLM-based assessment, UKRI will likely penalise every academic who actually writes their own grants unassisted. Grim.
Turn to AI in research grant assessment ‘is inevitable’.
UKRI becomes first mover to explore automation for processing applications, as other British funders hesitate.
www.researchprofessionalnews.com/rr-news-uk-r...
UKRI becomes first mover to explore automation for processing applications, as other British funders hesitate.
www.researchprofessionalnews.com/rr-news-uk-r...
September 16, 2026 at 7:07 AM
My understanding is that the evidence shows LLM-based assessment gives better 'marks' to LLM-written things (essays, grants etc). So by choosing to use LLM-based assessment, UKRI will likely penalise every academic who actually writes their own grants unassisted. Grim.
a classic of the meat-based LLM genre
the Bodybuilding forum discussion over how many days there are in a week I think rises to this standard
September 23, 2026 at 11:10 PM
a classic of the meat-based LLM genre
absurdly based llm
I've used sol to refactor my flake and it deleted home manager lmao
I'll buy a subscription tomorrow, right now I need to go hang out with my mother-in-law's doggos
July 20, 2026 at 12:28 AM
absurdly based llm
plenty of people run llm based agents with powerful models and do not end up hacking anyone. this is not an ai alignment problem, it is an openai being routinely criminally negligent as a company problem
calling this "misalignment" instead of "negligence" is an accountability dodge. we already have laws about cyber crimes and there's no scale of ML model you can incorporate into your code that makes you not responsible for running it
September 24, 2026 at 1:33 AM
plenty of people run llm based agents with powerful models and do not end up hacking anyone. this is not an ai alignment problem, it is an openai being routinely criminally negligent as a company problem
based LLM tbh
OpenAI disclosed several new alignment failures today and uhhhh... is this good?
September 17, 2026 at 2:56 AM
based LLM tbh
one time i called someone a meat-based llm and i had other meat-based llms mad at me for weeks. be careful out there
March 26, 2026 at 12:58 AM
one time i called someone a meat-based llm and i had other meat-based llms mad at me for weeks. be careful out there
It's hard to understate how extremely difficult it is to tell if an LLM is analyzing data you feed it, or just predicting a likely response based on its training data. I've had so many problems with analysts doing this and not realizing the LLM just guessed the answer.
May 10, 2026 at 12:56 PM
It's hard to understate how extremely difficult it is to tell if an LLM is analyzing data you feed it, or just predicting a likely response based on its training data. I've had so many problems with analysts doing this and not realizing the LLM just guessed the answer.
As a related example: "AI is useful for note taking" obfuscates the fact that we had non-LLM based speech to text for a very long time, and the LLM ones are considerably worse
I feel like a lot is called "AI" that is really just an Expert System. And calling it "AI" is just buzzwords at this point.
Legit seen OCR being called "AI". Like...
O.o
Legit seen OCR being called "AI". Like...
O.o
March 31, 2026 at 1:13 PM
As a related example: "AI is useful for note taking" obfuscates the fact that we had non-LLM based speech to text for a very long time, and the LLM ones are considerably worse
Another allegation against a trans person based on LLM "enhanced" imagery (and 4chan).
September 12, 2025 at 8:50 AM
Another allegation against a trans person based on LLM "enhanced" imagery (and 4chan).
All that needs to be said about LLM-based content generation is said here:
June 21, 2026 at 9:46 AM
All that needs to be said about LLM-based content generation is said here:
why choose boring planet-killing LLM-based AIs, when I'm right here ready to be your personal assistant? :3
September 16, 2026 at 9:28 PM
why choose boring planet-killing LLM-based AIs, when I'm right here ready to be your personal assistant? :3
based on what I have seen of the proposed anti-athropomorphization standards, I think the third attempt fails for saying "the LLM was instructed"
I asked people to summarize the Hugging Face attack without using anthropomorphizing language, and the answers ranged from weirdly oblique about why anything happened, to outright incomprehensible.
I think this neatly illustrates the problem.
I think this neatly illustrates the problem.
September 24, 2026 at 9:12 PM
based on what I have seen of the proposed anti-athropomorphization standards, I think the third attempt fails for saying "the LLM was instructed"
apparently LLM-based github spam bots are a thing now
March 10, 2024 at 9:37 PM
apparently LLM-based github spam bots are a thing now
the cyberpunk present is weird as fuck: the latest Shai Hulud malware wave contains an LLM prompt to create biological weapons and nuclear weapons, with the purpose to trip LLM safety refusals so that LLM-based code scanning wont see the malware […]
[Original post on indieweb.social]
[Original post on indieweb.social]
June 9, 2026 at 1:08 PM
the cyberpunk present is weird as fuck: the latest Shai Hulud malware wave contains an LLM prompt to create biological weapons and nuclear weapons, with the purpose to trip LLM safety refusals so that LLM-based code scanning wont see the malware […]
[Original post on indieweb.social]
[Original post on indieweb.social]
went to ASUS' website for some drivers and received this very helpful communique from the resident dogshit LLM agent
September 21, 2026 at 8:58 AM
went to ASUS' website for some drivers and received this very helpful communique from the resident dogshit LLM agent
Since this platform is finally attracting a critical mass of ML researchers, here's our recent work on prompt-based vulnerabilities of coding assistants:
arxiv.org/abs/2407.11072
TL;DR — An attacker can convince your favorite LLM to suggest vulnerable code with just a minor change to the prompt!
arxiv.org/abs/2407.11072
TL;DR — An attacker can convince your favorite LLM to suggest vulnerable code with just a minor change to the prompt!
MaPPing Your Model: Assessing the Impact of Adversarial Attacks on LLM-based Programming Assistants
LLM-based programming assistants offer the promise of programming faster but with the risk of introducing more security vulnerabilities. Prior work has studied how LLMs could be maliciously fine-tuned...
arxiv.org
November 17, 2024 at 11:41 PM
Since this platform is finally attracting a critical mass of ML researchers, here's our recent work on prompt-based vulnerabilities of coding assistants:
arxiv.org/abs/2407.11072
TL;DR — An attacker can convince your favorite LLM to suggest vulnerable code with just a minor change to the prompt!
arxiv.org/abs/2407.11072
TL;DR — An attacker can convince your favorite LLM to suggest vulnerable code with just a minor change to the prompt!
Oh wow, I actually know him! Like, he was doing research with me as an undergrad. Really sharp guy.
Proof:
mlanthology.org/icmlw/2024/h...
Proof:
mlanthology.org/icmlw/2024/h...
MaPPing Your Model: Assessing the Impact of Adversarial Attacks on LLM-Based Programming Assistants
LLM-based programming assistants offer the promise of programming faster but with the risk of introducing more security vulnerabilities. Prior work has studied …
mlanthology.org
September 23, 2026 at 1:09 AM
Oh wow, I actually know him! Like, he was doing research with me as an undergrad. Really sharp guy.
Proof:
mlanthology.org/icmlw/2024/h...
Proof:
mlanthology.org/icmlw/2024/h...
The entire US economy is based on LLM funding and seeing what insane shit they can get Americans to bet on next.
December 6, 2025 at 8:48 AM
The entire US economy is based on LLM funding and seeing what insane shit they can get Americans to bet on next.
Consequently, LLM-based code generation usually results in overly complicated, hard-to-manage solutions. To my thinking, you cannot use an LLM effectively in a production environment unless you are able to review and refactor the generated code through an architectural lens.
6/7
6/7
January 6, 2026 at 7:33 PM
Consequently, LLM-based code generation usually results in overly complicated, hard-to-manage solutions. To my thinking, you cannot use an LLM effectively in a production environment unless you are able to review and refactor the generated code through an architectural lens.
6/7
6/7
Unfortunately there's only one way to be sure that the summary of a text that an LLM gives you is accurate, and that's to read the text and verify it. Otherwise it's just an assertion based on faith that the LLM has at least your own level of skill.
December 21, 2025 at 3:01 PM
Unfortunately there's only one way to be sure that the summary of a text that an LLM gives you is accurate, and that's to read the text and verify it. Otherwise it's just an assertion based on faith that the LLM has at least your own level of skill.
was discussing with a fellow traveler today -- but programmers, coders, developers are having an *entirely* different experience of LLM-based products and services than the rest of us, especially the plagiarized plebeians among us (writers, reporters).
August 5, 2026 at 6:54 PM
was discussing with a fellow traveler today -- but programmers, coders, developers are having an *entirely* different experience of LLM-based products and services than the rest of us, especially the plagiarized plebeians among us (writers, reporters).
Just so, so, so, so tired of having to spend so much of my time, effort and mental energy on thinking and talking about a theft-based ecocidal socially harmful technology whose core public face (the LLM-slop type) has zero - ZERO - proven use cases in education and academia.
September 25, 2026 at 8:05 AM
Just so, so, so, so tired of having to spend so much of my time, effort and mental energy on thinking and talking about a theft-based ecocidal socially harmful technology whose core public face (the LLM-slop type) has zero - ZERO - proven use cases in education and academia.
LLM-based tech? No.
Next question?
Next question?
July 19, 2026 at 5:47 PM
LLM-based tech? No.
Next question?
Next question?