Ian Bicking
@ianbicking.org
Software developer in Minneapolis. Working with applications of LLMs. Previously: Mozilla, Meta, Brilliant.org
Excellent use of a Shepard tone metaphor
claim
from my reading of history the past was basically just worse in every way and median person got completely wrecked by things that don't really happen to ANYONE now, and a lot of present whining is just the bitter fruit of shepard tone type endless renormalization. but can I prove this? No
from my reading of history the past was basically just worse in every way and median person got completely wrecked by things that don't really happen to ANYONE now, and a lot of present whining is just the bitter fruit of shepard tone type endless renormalization. but can I prove this? No
I think this is true in the way that “industrialization enabled more harms than benefits” is true - society has not appropriately integrated world-breaking technological changes into the a pattern of living that prioritizes the median individual
December 10, 2025 at 2:08 AM
Excellent use of a Shepard tone metaphor
I feel similarly. Most AI is behavioral and it acts to diminish us and remove our autonomy. That is, it analyzes and manipulates our behavior, ignoring our intention and cognition. LLMs engage directly with those parts of ourselves.
the more I think about it, the more I think I unironically believe that if you are concerned about AI safety, you should be an LLM accelerationist
like, if you spend 15 minutes trying to imagine any other path to AI, their safety properties usually seem way worse
like, if you spend 15 minutes trying to imagine any other path to AI, their safety properties usually seem way worse
December 7, 2025 at 2:53 AM
I feel similarly. Most AI is behavioral and it acts to diminish us and remove our autonomy. That is, it analyzes and manipulates our behavior, ignoring our intention and cognition. LLMs engage directly with those parts of ourselves.
Reposted by Ian Bicking
I thought that Claude Code Skills would basically be bits of instructions that were loaded in automatically at the right time, but instead they seem to be explicitly loaded in by the agent...? This seems no better than files in docs/ ...?
November 29, 2025 at 9:00 PM
I thought that Claude Code Skills would basically be bits of instructions that were loaded in automatically at the right time, but instead they seem to be explicitly loaded in by the agent...? This seems no better than files in docs/ ...?
Claude Code has its own self-documentation index that it loads: code.claude.com/docs/en/clau...
Using Markdown files with links reminds me of Gopher or the Gemini Protocol (geminiprotocol.net)
Using Markdown files with links reminds me of Gopher or the Gemini Protocol (geminiprotocol.net)
code.claude.com
November 26, 2025 at 3:22 AM
Claude Code has its own self-documentation index that it loads: code.claude.com/docs/en/clau...
Using Markdown files with links reminds me of Gopher or the Gemini Protocol (geminiprotocol.net)
Using Markdown files with links reminds me of Gopher or the Gemini Protocol (geminiprotocol.net)
Reposted by Ian Bicking
Btw the word that’s currently in my display name, sphexish, was coined by AI researcher Douglas Hofstadter to mean “behaviors that appear conscious but are in fact preprogrammed”
It is taken from the Sphex wasp which exhibits some complex behaviors - which are completely programmed into its genes
It is taken from the Sphex wasp which exhibits some complex behaviors - which are completely programmed into its genes
November 25, 2025 at 9:14 PM
Btw the word that’s currently in my display name, sphexish, was coined by AI researcher Douglas Hofstadter to mean “behaviors that appear conscious but are in fact preprogrammed”
It is taken from the Sphex wasp which exhibits some complex behaviors - which are completely programmed into its genes
It is taken from the Sphex wasp which exhibits some complex behaviors - which are completely programmed into its genes
Opus in Claude Code is very eager to just do stuff without asking. Not sure how I feel about it yet...
But it's good Claude Code lets you add a message to the stream at any time, not having to wait for the agent to complete. It seems like an easy feature to implement, surprising other's haven't
But it's good Claude Code lets you add a message to the stream at any time, not having to wait for the agent to complete. It seems like an easy feature to implement, surprising other's haven't
November 25, 2025 at 8:01 PM
Opus in Claude Code is very eager to just do stuff without asking. Not sure how I feel about it yet...
But it's good Claude Code lets you add a message to the stream at any time, not having to wait for the agent to complete. It seems like an easy feature to implement, surprising other's haven't
But it's good Claude Code lets you add a message to the stream at any time, not having to wait for the agent to complete. It seems like an easy feature to implement, surprising other's haven't
I've reached the point on an almost-entirely vibecoded project where I need to do some major architectural refinement. Some thoughts...
November 20, 2025 at 12:29 AM
I've reached the point on an almost-entirely vibecoded project where I need to do some major architectural refinement. Some thoughts...
If you have kids (in your life) I will note that Shel Silverstein’s albums (A Light In The Attic and Where The Sidewalk Ends) are probably streaming on your service of choice (where you might not expect them) and they are delightful
November 19, 2025 at 2:33 PM
If you have kids (in your life) I will note that Shel Silverstein’s albums (A Light In The Attic and Where The Sidewalk Ends) are probably streaming on your service of choice (where you might not expect them) and they are delightful
Reposted by Ian Bicking
In About, With, Through, Without, Against: Five Ways to Learn AI, Leon Furze pushes back against simplistic binaries about whether AI helps or harms learning. #ArtificialIntelligence #AIEduf="/hashtag/AIEducation" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link">#AIEducation #AIEdu #AIInEd #AIInEdu #Podcast
About, With, Through, Without, Against: Five Ways to Learn AI
In About, With, Through, Without, Against: Five Ways to Learn AI, Leon Furze pushes back against simplistic binaries about whether AI helps or harms learning. He argues that young people are adaptable and capable, and proposes five overlapping approaches to thinking about AI in education. Learning about AI covers AI literacy—how to use and understand the technology—but Furze warns against making every teacher responsible for teaching AI skills. Learning with AI includes tutor chatbots and using AI for feedback, though he remains skeptical about personalized learning claims. Learning through AI treats the technology as a conduit or medium, like semantic search engines.
leonfurze.com
November 17, 2025 at 8:00 PM
In About, With, Through, Without, Against: Five Ways to Learn AI, Leon Furze pushes back against simplistic binaries about whether AI helps or harms learning. #ArtificialIntelligence #AIEduf="/hashtag/AIEducation" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link">#AIEducation #AIEdu #AIInEd #AIInEdu #Podcast
I suppose this is an obvious little guardrail for agentic coding, but I notice Claude Code isn't allowed to write to an existing file if it hasn't read the file first.
Soon we'll have something like a Terms Of Service for the AI that won't let you accept until you scroll to the bottom
Soon we'll have something like a Terms Of Service for the AI that won't let you accept until you scroll to the bottom
November 17, 2025 at 7:02 PM
I suppose this is an obvious little guardrail for agentic coding, but I notice Claude Code isn't allowed to write to an existing file if it hasn't read the file first.
Soon we'll have something like a Terms Of Service for the AI that won't let you accept until you scroll to the bottom
Soon we'll have something like a Terms Of Service for the AI that won't let you accept until you scroll to the bottom
I will counter: AI writing can be good, right now.
Today’s best models are sufficiently instructable to escape mid writing, but they won’t do it spontaneously. It also requires a framework, not just an initial prompt.
Today’s best models are sufficiently instructable to escape mid writing, but they won’t do it spontaneously. It also requires a framework, not just an initial prompt.
Why AI writing is mid
How the current way of training language models destroys any voice (and hope of good writing).
www.interconnects.ai/p/why-ai-wri...
How the current way of training language models destroys any voice (and hope of good writing).
www.interconnects.ai/p/why-ai-wri...
Why AI writing is mid
How the current way of training language models destroys any voice (and hope of good writing).
www.interconnects.ai
November 17, 2025 at 6:29 PM
I will counter: AI writing can be good, right now.
Today’s best models are sufficiently instructable to escape mid writing, but they won’t do it spontaneously. It also requires a framework, not just an initial prompt.
Today’s best models are sufficiently instructable to escape mid writing, but they won’t do it spontaneously. It also requires a framework, not just an initial prompt.
I was late to notice the $250 of free credits for Claude Code web, so I only have two days to use them. Honestly I'm finding it kind of annoying and unreliable compared to local usage, but dammit, I'm going to use those credits.
November 17, 2025 at 2:08 AM
I was late to notice the $250 of free credits for Claude Code web, so I only have two days to use them. Honestly I'm finding it kind of annoying and unreliable compared to local usage, but dammit, I'm going to use those credits.
I think I’d enjoy AI art presented in a much more deliberate way:
1. Describe what you are trying to achieve
2. Show the instantiation of the art
3. Reflect on the concrete elements that did or did not succeed
Simply rolling the aesthetic feedback dice gets tiring
1. Describe what you are trying to achieve
2. Show the instantiation of the art
3. Reflect on the concrete elements that did or did not succeed
Simply rolling the aesthetic feedback dice gets tiring
November 1, 2025 at 10:05 PM
I think I’d enjoy AI art presented in a much more deliberate way:
1. Describe what you are trying to achieve
2. Show the instantiation of the art
3. Reflect on the concrete elements that did or did not succeed
Simply rolling the aesthetic feedback dice gets tiring
1. Describe what you are trying to achieve
2. Show the instantiation of the art
3. Reflect on the concrete elements that did or did not succeed
Simply rolling the aesthetic feedback dice gets tiring
Reposted by Ian Bicking
[1/9] Excited to share our new paper "A Pragmatic View of AI Personhood" published today. We feel this topic is timely, and rapidly growing in importance as AI becomes agentic, as AI agents integrate further into the economy, and as more and more users encounter AI.
October 31, 2025 at 12:32 PM
[1/9] Excited to share our new paper "A Pragmatic View of AI Personhood" published today. We feel this topic is timely, and rapidly growing in importance as AI becomes agentic, as AI agents integrate further into the economy, and as more and more users encounter AI.
Thinking a little more about LLM sycophancy...
In general discernment is very hard to get right. You ask for critique and you'll get critique. You ask for a compliment and you'll get a compliment. There is no "just tell me the truth."
In general discernment is very hard to get right. You ask for critique and you'll get critique. You ask for a compliment and you'll get a compliment. There is no "just tell me the truth."
October 30, 2025 at 6:35 PM
Thinking a little more about LLM sycophancy...
In general discernment is very hard to get right. You ask for critique and you'll get critique. You ask for a compliment and you'll get a compliment. There is no "just tell me the truth."
In general discernment is very hard to get right. You ask for critique and you'll get critique. You ask for a compliment and you'll get a compliment. There is no "just tell me the truth."
I didn't think I was even using Claude Code that much, and I hit a rate limit. The rate limit also blocks me from using the normal Claude chat interface, which is an interesting choice. OpenAI's Codex similarly conked out fairly early with a rate limit, meanwhile Cursor keeps going and going...
October 28, 2025 at 9:57 PM
I didn't think I was even using Claude Code that much, and I hit a rate limit. The rate limit also blocks me from using the normal Claude chat interface, which is an interesting choice. OpenAI's Codex similarly conked out fairly early with a rate limit, meanwhile Cursor keeps going and going...
Reposted by Ian Bicking
TLDR; The PSF has made the decision to put our community and our shared diversity, equity, and inclusion values ahead of seeking $1.5M in new revenue. Please read and share. pyfound.blogspot.com/2025/10/NSF-...
🧵
🧵
The official home of the Python Programming Language
www.python.org
October 27, 2025 at 2:47 PM
TLDR; The PSF has made the decision to put our community and our shared diversity, equity, and inclusion values ahead of seeking $1.5M in new revenue. Please read and share. pyfound.blogspot.com/2025/10/NSF-...
🧵
🧵
I was reading this post talking about the great things about text adventures, and it got me thinking about parsing and representation... entropicthoughts.com/the-greatnes...
Specifically from the question: do text adventures really give players a lot of freedom?
Specifically from the question: do text adventures really give players a lot of freedom?
The Greatness of Text Adventures
entropicthoughts.com
October 22, 2025 at 6:26 PM
I was reading this post talking about the great things about text adventures, and it got me thinking about parsing and representation... entropicthoughts.com/the-greatnes...
Specifically from the question: do text adventures really give players a lot of freedom?
Specifically from the question: do text adventures really give players a lot of freedom?
In the small parking lot behind my house one of the employees always parks in the same place, with the placards “if you’re going to ride my ass at least pull my hair” and “bimbo on board” very visible in her back window. I admire the confidence!
October 17, 2025 at 4:08 PM
In the small parking lot behind my house one of the employees always parks in the same place, with the placards “if you’re going to ride my ass at least pull my hair” and “bimbo on board” very visible in her back window. I admire the confidence!
Reposted by Ian Bicking
If you want to give yourself nightmares about the Titan submersible disaster, this audio from the meeting where OceanGate fired the whistleblower bringing up safety concerns is Some Shit
October 16, 2025 at 11:31 PM
If you want to give yourself nightmares about the Titan submersible disaster, this audio from the meeting where OceanGate fired the whistleblower bringing up safety concerns is Some Shit
I did a little build-an-MCP server workshop for a local group I attend. Last time we looked it wasn’t easy to find MCP clients. With ChatGPT (paid) supporting MCP I thought it would work. It took an hour, but everyone who had the prerequisites did get there!
October 16, 2025 at 2:10 AM
I did a little build-an-MCP server workshop for a local group I attend. Last time we looked it wasn’t easy to find MCP clients. With ChatGPT (paid) supporting MCP I thought it would work. It took an hour, but everyone who had the prerequisites did get there!
Hot take: LLM agents could usually be a workflows (well defined pipelines), but letting the LLM take the wheel because you're too lazy to figure out what you're actually trying to do.
October 13, 2025 at 1:39 AM
Hot take: LLM agents could usually be a workflows (well defined pipelines), but letting the LLM take the wheel because you're too lazy to figure out what you're actually trying to do.
Having an iPhone after using Android for a long time, I now unconsciously notice the green vs. blue messages in iMessage, and I actively resent that they are taking that basic communication and applying a passive-aggressive branding.
October 13, 2025 at 1:07 AM
Having an iPhone after using Android for a long time, I now unconsciously notice the green vs. blue messages in iMessage, and I actively resent that they are taking that basic communication and applying a passive-aggressive branding.
I use Tools with LLMs for updating and managing state, and I’m realizing they aren’t a good fit. Tools seem biased toward querying and actions/verbs, not simple updates.
Most directly updates can be assumed to succeed, but Tools are expected to execute synchronously in the chat…
Most directly updates can be assumed to succeed, but Tools are expected to execute synchronously in the chat…
October 8, 2025 at 9:53 PM
I use Tools with LLMs for updating and managing state, and I’m realizing they aren’t a good fit. Tools seem biased toward querying and actions/verbs, not simple updates.
Most directly updates can be assumed to succeed, but Tools are expected to execute synchronously in the chat…
Most directly updates can be assumed to succeed, but Tools are expected to execute synchronously in the chat…