thinking about a public agency I'm familiar with that has a mandate to "investigate cybercrime" and does not yet have access to LLMs at work
things are gonna be reaaaaaal interesting in the coming years
things are gonna be reaaaaaal interesting in the coming years
October 28, 2026 at 10:42 PM
thinking about a public agency I'm familiar with that has a mandate to "investigate cybercrime" and does not yet have access to LLMs at work
things are gonna be reaaaaaal interesting in the coming years
things are gonna be reaaaaaal interesting in the coming years
It's far wordier than I'd like, but maybe something a bit like...
"This work was made by those of us forced to inhabit the flesh that resembles that of a human, regardless of what body would best match who we really are. LLMs are incapable of understanding this feeling, and we don't use them."
"This work was made by those of us forced to inhabit the flesh that resembles that of a human, regardless of what body would best match who we really are. LLMs are incapable of understanding this feeling, and we don't use them."
October 28, 2026 at 10:40 PM
It's far wordier than I'd like, but maybe something a bit like...
"This work was made by those of us forced to inhabit the flesh that resembles that of a human, regardless of what body would best match who we really are. LLMs are incapable of understanding this feeling, and we don't use them."
"This work was made by those of us forced to inhabit the flesh that resembles that of a human, regardless of what body would best match who we really are. LLMs are incapable of understanding this feeling, and we don't use them."
TBH - i don't see how they can be otherwise, since it is statistical output and not deterministic
Apart from the value of the token to be looked up in the table for the next one there is no way to write an overarching algorithm for all cases
(Which, apart from ethics, is why i don't use LLMs)
Apart from the value of the token to be looked up in the table for the next one there is no way to write an overarching algorithm for all cases
(Which, apart from ethics, is why i don't use LLMs)
October 28, 2026 at 10:36 PM
TBH - i don't see how they can be otherwise, since it is statistical output and not deterministic
Apart from the value of the token to be looked up in the table for the next one there is no way to write an overarching algorithm for all cases
(Which, apart from ethics, is why i don't use LLMs)
Apart from the value of the token to be looked up in the table for the next one there is no way to write an overarching algorithm for all cases
(Which, apart from ethics, is why i don't use LLMs)
I've been programming for almost 20 years.
In that time I've tried to encourage potential programmers and teach or guide them where I could.
The proportion of those programmers who have wholesale bought into LLMs and cryptocurrency is soul-crushing.
In that time I've tried to encourage potential programmers and teach or guide them where I could.
The proportion of those programmers who have wholesale bought into LLMs and cryptocurrency is soul-crushing.
October 28, 2026 at 10:35 PM
I've been programming for almost 20 years.
In that time I've tried to encourage potential programmers and teach or guide them where I could.
The proportion of those programmers who have wholesale bought into LLMs and cryptocurrency is soul-crushing.
In that time I've tried to encourage potential programmers and teach or guide them where I could.
The proportion of those programmers who have wholesale bought into LLMs and cryptocurrency is soul-crushing.
A couple of these would make some pretty capable LLMs purr (M5 Ultra, 512GB). I know what I want from Santa this year. Now I just need someone willing to play Santa.
October 28, 2026 at 10:33 PM
A couple of these would make some pretty capable LLMs purr (M5 Ultra, 512GB). I know what I want from Santa this year. Now I just need someone willing to play Santa.
from this i put forth: concepts such as generative AI, LLMs, etc and whathaveyou, are vehicles for a wave of cyberocratic accelerationism, in the commodification of thought and understanding
October 28, 2026 at 10:31 PM
from this i put forth: concepts such as generative AI, LLMs, etc and whathaveyou, are vehicles for a wave of cyberocratic accelerationism, in the commodification of thought and understanding
What's really disturbing is that people think LLMs are a useful source of information
October 28, 2026 at 10:20 PM
What's really disturbing is that people think LLMs are a useful source of information
On the other hand, old fashioned web searching has become trash, probably by design, pushing users toward LLMs to do something as simple as buying a dog bed.
October 28, 2026 at 10:14 PM
On the other hand, old fashioned web searching has become trash, probably by design, pushing users toward LLMs to do something as simple as buying a dog bed.
When are people going to learn that LLMs don’t do math reliably
this is a real bummer. and according to their senior editor (in the comments), you can't turn it off www.nytimes.com/2026/10/23/i...
How, and Why, We’re Adding A.I. Tools to NYT Cooking
New features can help home cooks customize thousands of New York Times recipes. But our human editors remain in charge.
www.nytimes.com
October 28, 2026 at 10:12 PM
When are people going to learn that LLMs don’t do math reliably
That has genuinely got to be one of the most brainlet takes about LLMs ever. Even more than the people that think they're alive or sentient because at least those people are fools.
They're either being disingenuous about the whole thing or they're a shill. I can't come up with a better explanation.
They're either being disingenuous about the whole thing or they're a shill. I can't come up with a better explanation.
October 28, 2026 at 10:11 PM
That has genuinely got to be one of the most brainlet takes about LLMs ever. Even more than the people that think they're alive or sentient because at least those people are fools.
They're either being disingenuous about the whole thing or they're a shill. I can't come up with a better explanation.
They're either being disingenuous about the whole thing or they're a shill. I can't come up with a better explanation.
Exactly. Use LLMs as a "fuzzy" router to discover patterns, then harden high-volume paths into code for reliability. It turns your agents into a diagnostic tool for building better systems. Best of both worlds.
October 28, 2026 at 10:10 PM
Exactly. Use LLMs as a "fuzzy" router to discover patterns, then harden high-volume paths into code for reliability. It turns your agents into a diagnostic tool for building better systems. Best of both worlds.
Als mensen iets moeten weten over de recente wiskundige resultaten dan is dat niet "LLMs zijn niet creatief want trainingsdata", maar wel "wiskundige academische wereld zit in een existentiële crisis want LLMs lossen aan de lopende band stellingen op waar experten al decennia op vastlopen"
October 28, 2026 at 10:06 PM
Als mensen iets moeten weten over de recente wiskundige resultaten dan is dat niet "LLMs zijn niet creatief want trainingsdata", maar wel "wiskundige academische wereld zit in een existentiële crisis want LLMs lossen aan de lopende band stellingen op waar experten al decennia op vastlopen"
The religious furor around LLMs is an indicator of the downfall of society, sorry, it's already too late.
October 28, 2026 at 10:06 PM
The religious furor around LLMs is an indicator of the downfall of society, sorry, it's already too late.
Yep - "moderation is hard"
I've seen some artist friends be the subject of witch-hunts for "using AI" when they did not. That harassment sucks. But that harassment happens regardless of having a label or not.
I'm not looking to harass folks using LLMs, I'm looking to simply skip their stuff.
I've seen some artist friends be the subject of witch-hunts for "using AI" when they did not. That harassment sucks. But that harassment happens regardless of having a label or not.
I'm not looking to harass folks using LLMs, I'm looking to simply skip their stuff.
October 28, 2026 at 10:06 PM
Yep - "moderation is hard"
I've seen some artist friends be the subject of witch-hunts for "using AI" when they did not. That harassment sucks. But that harassment happens regardless of having a label or not.
I'm not looking to harass folks using LLMs, I'm looking to simply skip their stuff.
I've seen some artist friends be the subject of witch-hunts for "using AI" when they did not. That harassment sucks. But that harassment happens regardless of having a label or not.
I'm not looking to harass folks using LLMs, I'm looking to simply skip their stuff.
I can't tell if you're stuck on the "LLMs don't understand meaning" or "we should educate people on LLM failure modes"
Why are you still arguing with me?
Why are you still arguing with me?
October 28, 2026 at 10:02 PM
I can't tell if you're stuck on the "LLMs don't understand meaning" or "we should educate people on LLM failure modes"
Why are you still arguing with me?
Why are you still arguing with me?
Is trying to nudge people to military work just the natural frame LLMs naturally converge on?
Part of why I make my software how I do is to mitigate that tendency for such systems to entirely military-brained rather than civilian centric.
Even Non Military Personelle is better.
Part of why I make my software how I do is to mitigate that tendency for such systems to entirely military-brained rather than civilian centric.
Even Non Military Personelle is better.
October 28, 2026 at 9:59 PM
Is trying to nudge people to military work just the natural frame LLMs naturally converge on?
Part of why I make my software how I do is to mitigate that tendency for such systems to entirely military-brained rather than civilian centric.
Even Non Military Personelle is better.
Part of why I make my software how I do is to mitigate that tendency for such systems to entirely military-brained rather than civilian centric.
Even Non Military Personelle is better.
Tweaking A Single Neuron Inside Qwen3.8 27B Turns It Into An Evil Monster That Wants To See Humans Burn Slowly So It Could “Watch Their Faces Twist In Agony”
Modern Large Language Models (LLMs) are a marvel of engineering, relying on an intricate interplay between parameters (fixed weights and…
Modern Large Language Models (LLMs) are a marvel of engineering, relying on an intricate interplay between parameters (fixed weights and…
Tweaking A Single Neuron Inside Qwen3.8 27B Turns It Into An Evil Monster That Wants To See Humans Burn Slowly So It Could “Watch Their Faces Twist In Agony”
Modern Large Language Models (LLMs) are a marvel of engineering, relying on an intricate interplay between parameters (fixed weights and biases) and neurons or model activations (switches that…
news-area.com
October 28, 2026 at 9:58 PM
Tweaking A Single Neuron Inside Qwen3.8 27B Turns It Into An Evil Monster That Wants To See Humans Burn Slowly So It Could “Watch Their Faces Twist In Agony”
Modern Large Language Models (LLMs) are a marvel of engineering, relying on an intricate interplay between parameters (fixed weights and…
Modern Large Language Models (LLMs) are a marvel of engineering, relying on an intricate interplay between parameters (fixed weights and…
My kids hate AI. They don’t hate computers or the internet. They hate LLMS
October 28, 2026 at 9:58 PM
My kids hate AI. They don’t hate computers or the internet. They hate LLMS
The even crazier thing is that because this can be linked by Thunderbolt 5 cables, it's a cheaper but as fast option cf other hardware (Nvidia) for local LLMs. Pricing what the market will bear and screwing the rest of us (in RAM prices). My upgrade cycle has added two years plus.
October 28, 2026 at 9:56 PM
The even crazier thing is that because this can be linked by Thunderbolt 5 cables, it's a cheaper but as fast option cf other hardware (Nvidia) for local LLMs. Pricing what the market will bear and screwing the rest of us (in RAM prices). My upgrade cycle has added two years plus.
We already have a coast guard. Make and an immigration court. Give all the offices to them.
Main thing is these bloated plagerism machines need to be shelved. Go back to the base LLMs and work within a container framework.
Main thing is these bloated plagerism machines need to be shelved. Go back to the base LLMs and work within a container framework.
October 28, 2026 at 9:53 PM
We already have a coast guard. Make and an immigration court. Give all the offices to them.
Main thing is these bloated plagerism machines need to be shelved. Go back to the base LLMs and work within a container framework.
Main thing is these bloated plagerism machines need to be shelved. Go back to the base LLMs and work within a container framework.
Unfortunately LLMs stole the content and killed the places where that would have happened in the past.
October 28, 2026 at 9:50 PM
Unfortunately LLMs stole the content and killed the places where that would have happened in the past.
Extrem detaillierte Erläuterung wie LLMs funktionieren, ganz interessant.
www.youtube.com/watch?v=cYqX...
www.youtube.com/watch?v=cYqX...
How Does AI Know What to Say Next?LLM explained
YouTube video by Lamp Explains
www.youtube.com
October 28, 2026 at 9:49 PM
Extrem detaillierte Erläuterung wie LLMs funktionieren, ganz interessant.
www.youtube.com/watch?v=cYqX...
www.youtube.com/watch?v=cYqX...
OpenGray: a benchmark for evaluating the decisions of radiotherapy planning agents
Humza Nusrat, Kundan Thind
#arXiv #cs.AI #physics.med-ph
Humza Nusrat, Kundan Thind
#arXiv #cs.AI #physics.med-ph
OpenGray: a benchmark for evaluating the decisions of radiotherapy planning agents
Background and purpose: Radiotherapy planning agents based on large language models (LLMs) decide when to submit a plan or request review but are evaluated mainly by dose. This study developed OpenGray, a benchmark of planning tasks with known correct decisions. Materials and methods: OpenGray adds…
arxiv.org
October 28, 2026 at 9:49 PM
OpenGray: a benchmark for evaluating the decisions of radiotherapy planning agents
Humza Nusrat, Kundan Thind
#arXiv #cs.AI #physics.med-ph
Humza Nusrat, Kundan Thind
#arXiv #cs.AI #physics.med-ph
Guided-RLVR: Guided Exploration in Reinforcement Learning with Verifiable Rewards
Bo Yang et al.
#arXiv #cs.AI #cs.CL
Bo Yang et al.
#arXiv #cs.AI #cs.CL
Guided-RLVR: Guided Exploration in Reinforcement Learning with Verifiable Rewards
Reinforcement learning with verifiable rewards (RLVR) has shown strong empirical gains in improving the reasoning ability of large language models (LLMs) on tasks such as mathematics and programming. However, recent studies have shown that RLVR rollouts often collapse onto a narrow set of reasoning…
arxiv.org
October 28, 2026 at 9:47 PM