I still believe it's a mistake to believe that the fact that LLM capability along certain dimensions seems to increase with scale implies that a sufficiently large LLM can do anything imaginable. Or that if an LLM can perform a task with difficulty D, it can do anything with difficulty ≤ D.
May 21, 2026 at 6:50 PM
I still believe it's a mistake to believe that the fact that LLM capability along certain dimensions seems to increase with scale implies that a sufficiently large LLM can do anything imaginable. Or that if an LLM can perform a task with difficulty D, it can do anything with difficulty ≤ D.
LLM chatbots have also been given a false “scientific” authority that your d&d friends and musicians don’t have
January 4, 2026 at 10:24 PM
LLM chatbots have also been given a false “scientific” authority that your d&d friends and musicians don’t have
A thing students still don't seem to understand about my GPT/LLM/"AI" policy— even though I tell them over and over and FUCKING OVER— is that I will know their undisclosed LLM usage by the trail of its bullshit.
So when I give you a D- about it, know that that's me being *kind*.
So when I give you a D- about it, know that that's me being *kind*.
December 15, 2024 at 2:52 AM
A thing students still don't seem to understand about my GPT/LLM/"AI" policy— even though I tell them over and over and FUCKING OVER— is that I will know their undisclosed LLM usage by the trail of its bullshit.
So when I give you a D- about it, know that that's me being *kind*.
So when I give you a D- about it, know that that's me being *kind*.
More recent job AI job listings by Mihoyo
March 5, 2026 at 11:27 PM
More recent job AI job listings by Mihoyo
i think LLMs are fascinating technology with a number of likely use cases when that technology is further refined, but it doesn’t really bode well that a not-insignificant number of enthusiasts and even pioneers in LLM R&D exhibit varying degrees of superstitious belief regarding their sentience
February 23, 2026 at 4:02 AM
i think LLMs are fascinating technology with a number of likely use cases when that technology is further refined, but it doesn’t really bode well that a not-insignificant number of enthusiasts and even pioneers in LLM R&D exhibit varying degrees of superstitious belief regarding their sentience
“Are you telling me the LLM is so spiked that all it can do now is put d cups on Benjamin Disraeli?!”
November 28, 2024 at 2:30 AM
“Are you telling me the LLM is so spiked that all it can do now is put d cups on Benjamin Disraeli?!”
June 5, 2025 at 7:33 AM
This is a really neat Python library providing decorator and docstring and type signature based syntactic sugar over my LLM package - I blogged about here: simonwillison.net/2025/Apr/3/s...
April 3, 2025 at 3:00 PM
This is a really neat Python library providing decorator and docstring and type signature based syntactic sugar over my LLM package - I blogged about here: simonwillison.net/2025/Apr/3/s...
The implication that installing the LLM program itself isn't immediately harmful when those programs can get access to your computer almost carte blanche if you simply toggle a switch is W I L D.
September 20, 2026 at 9:48 AM
The implication that installing the LLM program itself isn't immediately harmful when those programs can get access to your computer almost carte blanche if you simply toggle a switch is W I L D.
And things like 'cancer detection' are NOT the same as the destructive generative LLM systems. :D
February 12, 2026 at 1:46 AM
And things like 'cancer detection' are NOT the same as the destructive generative LLM systems. :D
Opus 4.8 (the very latest LLM from Anthropic) responds to a particularly tricky question, "How many days in the week have the letter "d" in them?"
AI is going to change EVERYTHING!
AI is going to change EVERYTHING!
![[Opus 4.8] "Two--Sunday and... wait, let me actually check.
Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, Sunday.
The ones with a "d": Wednesday, Thursday, and Sunday all have one. Monday too. So that's Monday, Tuesday, Wednesday, Thursday Sunday--four days.
(My first instinct was wrong, hence the correction.)](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:wstpte2vzb4ra2iqtnw2d2ef/bafkreifgj3dad6m5iow4kti6oojtjt7oe6l2wkdgw5mt7njvp33hl7n4lq)
May 31, 2026 at 12:13 PM
Opus 4.8 (the very latest LLM from Anthropic) responds to a particularly tricky question, "How many days in the week have the letter "d" in them?"
AI is going to change EVERYTHING!
AI is going to change EVERYTHING!
More indications I am an LLM, I see! :D
September 28, 2026 at 12:20 PM
More indications I am an LLM, I see! :D
“Are you telling me the LLM is so spiked that all it can do now is put d cups on Benjamin Disraeli?!" - @militantrabbit.bsky.social
November 28, 2024 at 2:51 AM
“Are you telling me the LLM is so spiked that all it can do now is put d cups on Benjamin Disraeli?!" - @militantrabbit.bsky.social
Ultimately, video games or heavy metal music or D&D are the works of people communicating ideas to one another. LLM chatbots are emphatically not that. They're an automated system that generates a slurry of input that resembles communication with another person, but isn't.
January 4, 2026 at 10:19 PM
Ultimately, video games or heavy metal music or D&D are the works of people communicating ideas to one another. LLM chatbots are emphatically not that. They're an automated system that generates a slurry of input that resembles communication with another person, but isn't.
Ye olde Seventh Sanctum. Saved me so much time by helping me settle on names for my D&D campaigns, and required exactly ZERO LLM.
www.seventhsanctum.com/index-name.php
www.seventhsanctum.com/index-name.php
Seventh Sanctum: Name Generators
Seventh Sanctum's generators for names both general and specific!
www.seventhsanctum.com
August 3, 2025 at 5:42 PM
Ye olde Seventh Sanctum. Saved me so much time by helping me settle on names for my D&D campaigns, and required exactly ZERO LLM.
www.seventhsanctum.com/index-name.php
www.seventhsanctum.com/index-name.php
On December 31st Mihoyo posted several new job listings, including a whole page of AI themed jobs.
Take a look.
Take a look.
January 7, 2026 at 9:07 PM
On December 31st Mihoyo posted several new job listings, including a whole page of AI themed jobs.
Take a look.
Take a look.
lmao literally right after i posted this they @'d everyone to announce they JUST released their new Paper Mario port, and the newly-made channel for it keeps having "was this made with any LLM tools?" messages get deleted within seconds. they had their mods READY for this shit. it's hilariously sad
man this is so deeply cringe, the SoH discussion channel in the HarborMasters discord is now full of ppl calling each other idiots for "not knowing the difference between vibe-coding and AI assistance" and calling them ungrateful cuz "people are volunteering their unpaid time so you can play this"
September 17, 2026 at 2:20 AM
lmao literally right after i posted this they @'d everyone to announce they JUST released their new Paper Mario port, and the newly-made channel for it keeps having "was this made with any LLM tools?" messages get deleted within seconds. they had their mods READY for this shit. it's hilariously sad
As someone who has professionally run thousands of hours of D&D for entertainment:
NPCs should be flavorful, memorable, and provide the player with what they need to have high agency. LLM AI chatbots have a hard time doing all three, especially over a duration.
NPCs should be flavorful, memorable, and provide the player with what they need to have high agency. LLM AI chatbots have a hard time doing all three, especially over a duration.
I'm annoyed by the AI NPC thing again. It's like saying "what if you could open every drawer" or "go into every building." Those things are only fun in games if there's something to find in there. If 80% of them are empty, then you've wasted my precious gaming time.
March 10, 2026 at 5:01 PM
As someone who has professionally run thousands of hours of D&D for entertainment:
NPCs should be flavorful, memorable, and provide the player with what they need to have high agency. LLM AI chatbots have a hard time doing all three, especially over a duration.
NPCs should be flavorful, memorable, and provide the player with what they need to have high agency. LLM AI chatbots have a hard time doing all three, especially over a duration.
LLM-famous is the new Wikipedia-famous. The less LLMs allucinate about you, the more Internet famous you are :D
January 25, 2025 at 12:53 PM
LLM-famous is the new Wikipedia-famous. The less LLMs allucinate about you, the more Internet famous you are :D
FWIW, I was trying to get some sleep; the claim that Trump's tariffs were LLM-generated crossed my mind, connected with the thought about people not having the knowledge to assess LLM output, which then connected with the D-K effect.
I'm sure someone smarter than me has already written about it.
I'm sure someone smarter than me has already written about it.
April 4, 2025 at 3:28 AM
FWIW, I was trying to get some sleep; the claim that Trump's tariffs were LLM-generated crossed my mind, connected with the thought about people not having the knowledge to assess LLM output, which then connected with the D-K effect.
I'm sure someone smarter than me has already written about it.
I'm sure someone smarter than me has already written about it.
Especially d) is quite profound - how did we get to the point where LLMs are SO heavily subsidized that devs can waste compute on using LLMs for tasks that (much more efficient) classical algorithms can handle.
The market is deeply unhealthy and it's time to stop these wasteful LLM subsidies.
The market is deeply unhealthy and it's time to stop these wasteful LLM subsidies.
Reminded of someone trying to "train" LLMs to do "sorting", to prove LLMs can "solve problems".
Not only a) are LLMs not reliable, but also b) sorting is a computationally easy to solve task, c) we have many efficient algorithms for it, and d) LLMs unnecessarily bloat the resource consumption.
Not only a) are LLMs not reliable, but also b) sorting is a computationally easy to solve task, c) we have many efficient algorithms for it, and d) LLMs unnecessarily bloat the resource consumption.
May 22, 2026 at 11:28 PM
Especially d) is quite profound - how did we get to the point where LLMs are SO heavily subsidized that devs can waste compute on using LLMs for tasks that (much more efficient) classical algorithms can handle.
The market is deeply unhealthy and it's time to stop these wasteful LLM subsidies.
The market is deeply unhealthy and it's time to stop these wasteful LLM subsidies.
Our answer: KServe + @llm-d.ai + vLLM with prefix-cache aware routing.
The results: 3x more output tokens/sec and 2x faster time to first token.
Check out the full writeup here: llm-d.ai/blog/product...
The results: 3x more output tokens/sec and 2x faster time to first token.
Check out the full writeup here: llm-d.ai/blog/product...
Production-Grade LLM Inference at Scale with KServe, llm-d, and vLLM | llm-d
How migrating from a simple vLLM deployment to a robust MLOps platform utilizing KServe, llm-d's intelligent routing, and vLLM solved significant scaling and operational challenges in LLM deployment t...
llm-d.ai
April 22, 2026 at 2:19 AM
Our answer: KServe + @llm-d.ai + vLLM with prefix-cache aware routing.
The results: 3x more output tokens/sec and 2x faster time to first token.
Check out the full writeup here: llm-d.ai/blog/product...
The results: 3x more output tokens/sec and 2x faster time to first token.
Check out the full writeup here: llm-d.ai/blog/product...
bin noch immer enttäuscht, dass die Grazer Linux Tage den Talk von @schenior.bsky.social und mir über LLM Nutzung bei österreichischen Medien klanglos abgelehnt haben.
Hier der volle (LLM-generierte) Bericht :D
sink.mariozechner.at/api/share/10...
Hier der volle (LLM-generierte) Bericht :D
sink.mariozechner.at/api/share/10...
Detection of Large Language Model Usage in Austrian News Media Through Linguistic Markers
sink.mariozechner.at
March 8, 2026 at 8:18 PM
bin noch immer enttäuscht, dass die Grazer Linux Tage den Talk von @schenior.bsky.social und mir über LLM Nutzung bei österreichischen Medien klanglos abgelehnt haben.
Hier der volle (LLM-generierte) Bericht :D
sink.mariozechner.at/api/share/10...
Hier der volle (LLM-generierte) Bericht :D
sink.mariozechner.at/api/share/10...
Are you sure you don't have a LLM running in background mining for lean proofs? :D
September 28, 2026 at 8:15 AM
Are you sure you don't have a LLM running in background mining for lean proofs? :D