iamwil
@interjectedfuture.com
Tech Zine Issue 1: LLM System Eval https://forestfriends.tech
Local-first/Reactive Programming ⁙ LLM system evals ⁙ Startup lessons ⁙ Game design quips.
Longform: https://interjectedfuture.com
Podcast: https://www.youtube.com/@techniumpod
Local-first/Reactive Programming ⁙ LLM system evals ⁙ Startup lessons ⁙ Game design quips.
Longform: https://interjectedfuture.com
Podcast: https://www.youtube.com/@techniumpod
Pinned
iamwil
@interjectedfuture.com
· Mar 22
I wish these end-of-the-year summaries with badges and cards were soul-bound NFTs; or some way to collect them over the years, even if the companies go out of business. It's a missing opportunity to tell people a story of their personal journey with a long term arc.
December 25, 2025 at 7:00 PM
I wish these end-of-the-year summaries with badges and cards were soul-bound NFTs; or some way to collect them over the years, even if the companies go out of business. It's a missing opportunity to tell people a story of their personal journey with a long term arc.
Reposted by iamwil
🧵 Could Bitchat have 5x the range and 100x the throughput for the same power expenditure? I explored how Wi-Fi Aware could improve the reliability and throughput of Bitchat and mobile ad-hoc networks in the absence of internet connectivity. #bitchat
👇
👇
December 23, 2025 at 4:09 PM
🧵 Could Bitchat have 5x the range and 100x the throughput for the same power expenditure? I explored how Wi-Fi Aware could improve the reliability and throughput of Bitchat and mobile ad-hoc networks in the absence of internet connectivity. #bitchat
👇
👇
No one wants to do the heavy lift of building offline-first or local-first into their product until their first hearing. Hearing-first software.
December 24, 2025 at 4:10 PM
No one wants to do the heavy lift of building offline-first or local-first into their product until their first hearing. Hearing-first software.
Software performance isn't everything. You have to balance time to delivery and cost. But we're so off the performance curve that I now have spinners for cutting and pasting text.
Once you see it, you see it everywhere.
Once you see it, you see it everywhere.
December 22, 2025 at 7:00 PM
Software performance isn't everything. You have to balance time to delivery and cost. But we're so off the performance curve that I now have spinners for cutting and pasting text.
Once you see it, you see it everywhere.
Once you see it, you see it everywhere.
Just as Wikipedia is a child of the Internet, the Primer should be a child of LLMs. I want my Illustrated Primer. But the optimism prevalent at the Internet Age is noticeably missing in the Intelligence Age. So we can't imagine it. And we can't build what we can't imagine.
December 20, 2025 at 4:00 PM
Just as Wikipedia is a child of the Internet, the Primer should be a child of LLMs. I want my Illustrated Primer. But the optimism prevalent at the Internet Age is noticeably missing in the Intelligence Age. So we can't imagine it. And we can't build what we can't imagine.
I don't mind what he's doing. But I want to know how an immortal population would work in a democracy. If it's true that old minds don't change their minds as quickly as young minds (if at all), then public policy would stagnate. And so would society.
December 18, 2025 at 8:00 PM
I don't mind what he's doing. But I want to know how an immortal population would work in a democracy. If it's true that old minds don't change their minds as quickly as young minds (if at all), then public policy would stagnate. And so would society.
I guess we're so used to tabs now I didn't notice this detail: A tab is a quick access to a page metaphored by physical tabs in a stack of paper. Hence, when viewing a page, the tag should looked connected to the page. But here, the tab is connected to the toolbar!
December 17, 2025 at 10:00 PM
I guess we're so used to tabs now I didn't notice this detail: A tab is a quick access to a page metaphored by physical tabs in a stack of paper. Hence, when viewing a page, the tag should looked connected to the page. But here, the tab is connected to the toolbar!
This was pretty great. I've always loved the explorable explainers, in the tradition of Bret Victor. However, they're expensive to produce. People ask: "What's after the chat interface for LLMs?"
Well, here's one answer: explorable explainers.
Well, here's one answer: explorable explainers.
December 17, 2025 at 8:00 PM
This was pretty great. I've always loved the explorable explainers, in the tradition of Bret Victor. However, they're expensive to produce. People ask: "What's after the chat interface for LLMs?"
Well, here's one answer: explorable explainers.
Well, here's one answer: explorable explainers.
The affordance for working with LLMs is the same as affordance for working with other humans.
December 16, 2025 at 8:57 PM
The affordance for working with LLMs is the same as affordance for working with other humans.
"How do we build programming with affordances for LLMs?" My answer usually is "the same affordances for humans", since LLMs are trained on human coding tools. But here's a different answer: expose intermediate declarative data, like the XML UI tree.
x.com/ethanjlim/s...
x.com/ethanjlim/s...
December 16, 2025 at 7:00 PM
"How do we build programming with affordances for LLMs?" My answer usually is "the same affordances for humans", since LLMs are trained on human coding tools. But here's a different answer: expose intermediate declarative data, like the XML UI tree.
x.com/ethanjlim/s...
x.com/ethanjlim/s...
Reposted by iamwil
Separating side effects from your code so AI can test it without lying or creating mock slop. interjectedfuture.com/what-is-alge...
What is Algebraic about Algebraic Effects?
Compositionality can be much more than just an interface between two objects or functions. It can be a set of laws.
interjectedfuture.com
December 15, 2025 at 5:13 PM
Separating side effects from your code so AI can test it without lying or creating mock slop. interjectedfuture.com/what-is-alge...
Maybe design docs keep getting written, though no one reads them, is because it's the default tool to assist with thinking.
But writing is a terrible medium for systems thinking. It can name all the parts, but not so much how they relate. It'd be too tedious.
But writing is a terrible medium for systems thinking. It can name all the parts, but not so much how they relate. It'd be too tedious.
December 15, 2025 at 8:58 PM
Maybe design docs keep getting written, though no one reads them, is because it's the default tool to assist with thinking.
But writing is a terrible medium for systems thinking. It can name all the parts, but not so much how they relate. It'd be too tedious.
But writing is a terrible medium for systems thinking. It can name all the parts, but not so much how they relate. It'd be too tedious.
Reposted by iamwil
all my homies are talking about the Lean prover. AoC in Lean, hardware design in Lean, you name it
learn Lean challenge 2026
learn Lean challenge 2026
This blog post interjectedfuture.com/the-best-way... on using LLMs to learn how to do formal proofs is another nice angle on the topic
December 13, 2025 at 11:39 AM
all my homies are talking about the Lean prover. AoC in Lean, hardware design in Lean, you name it
learn Lean challenge 2026
learn Lean challenge 2026
Directionally, I agree! We've already observed 1) LLMs remove activation barrier to other ecosystems. 2) LLMs remove the tactical tedium of a language. That means a wider range of tools are available to wield practically. But it also comes with caveats.
bsky.app/profile/mar...
bsky.app/profile/mar...
December 14, 2025 at 8:00 PM
Directionally, I agree! We've already observed 1) LLMs remove activation barrier to other ecosystems. 2) LLMs remove the tactical tedium of a language. That means a wider range of tools are available to wield practically. But it also comes with caveats.
bsky.app/profile/mar...
bsky.app/profile/mar...
A plea by Scott Jenson to start pushing the boundaries of human interface design on desktop again. An interesting idea is what he calls a loop, borrowed from game design, but in game design circles, it seems to be called depth.
youtu.be/1fZTOjd_bOQ...
youtu.be/1fZTOjd_bOQ...
Are we stuck with the same Desktop UX forever? | Ubuntu Summit 25.10
This talk focuses on that evil little term “UX/UI,” which is responsible for so much confusion and tension in open-source projects. Not only does it unnecess...
www.youtube.com
December 14, 2025 at 7:30 PM
A plea by Scott Jenson to start pushing the boundaries of human interface design on desktop again. An interesting idea is what he calls a loop, borrowed from game design, but in game design circles, it seems to be called depth.
youtu.be/1fZTOjd_bOQ...
youtu.be/1fZTOjd_bOQ...
Reposted by iamwil
This blog post interjectedfuture.com/the-best-way... on using LLMs to learn how to do formal proofs is another nice angle on the topic
December 13, 2025 at 10:57 AM
This blog post interjectedfuture.com/the-best-way... on using LLMs to learn how to do formal proofs is another nice angle on the topic
Sometimes, I wonder if this is where logic programming has an advantage for: "what kind of language is great for working with LLMs". You don't have to simulate state in your head when reviewing the code. And LLMs can run inference on the code they wrote to self-correct.
x.com/ankitiscrac...
x.com/ankitiscrac...
December 12, 2025 at 8:00 PM
Sometimes, I wonder if this is where logic programming has an advantage for: "what kind of language is great for working with LLMs". You don't have to simulate state in your head when reviewing the code. And LLMs can run inference on the code they wrote to self-correct.
x.com/ankitiscrac...
x.com/ankitiscrac...
I don't think managing multiple agents to vibe code is the future of engineering (nor is just hand writing everything either). Until AI can solve hard out-of-band problems, humans will have to do that work. And that work requires flow, which cannot happen when babysitting agents.
December 11, 2025 at 8:00 PM
I don't think managing multiple agents to vibe code is the future of engineering (nor is just hand writing everything either). Until AI can solve hard out-of-band problems, humans will have to do that work. And that work requires flow, which cannot happen when babysitting agents.
I've wondered if this idea can be extended: given how individual employees converse, think, write, and code, can we give a weight to their contribution to that B2B sale, all the way back to the academic? Can we "backprop" on this network, and adjust weights?
x.com/ID_AA_Carma...
x.com/ID_AA_Carma...
December 10, 2025 at 8:00 PM
I've wondered if this idea can be extended: given how individual employees converse, think, write, and code, can we give a weight to their contribution to that B2B sale, all the way back to the academic? Can we "backprop" on this network, and adjust weights?
x.com/ID_AA_Carma...
x.com/ID_AA_Carma...
The more nuanced and tacit your work, the less likely AI agents will do them to completion. So you end up baby-sitting & multi-tasking multiple agents, destroying your focus and flow.
x.com/arafatkatze...
x.com/arafatkatze...
December 9, 2025 at 8:00 PM
The more nuanced and tacit your work, the less likely AI agents will do them to completion. So you end up baby-sitting & multi-tasking multiple agents, destroying your focus and flow.
x.com/arafatkatze...
x.com/arafatkatze...
Maybe if top-of-the-line Nvidia chips were only priced in USD-backed stablecoins, it'd extend the petrodollar dominance of USD into the era of cryptodollar dominance of USD.
But unlike oil, this seems like a weaker chokehold.
But unlike oil, this seems like a weaker chokehold.
December 9, 2025 at 7:30 PM
Maybe if top-of-the-line Nvidia chips were only priced in USD-backed stablecoins, it'd extend the petrodollar dominance of USD into the era of cryptodollar dominance of USD.
But unlike oil, this seems like a weaker chokehold.
But unlike oil, this seems like a weaker chokehold.
Note-taking app ecosystems should run a collective decentralized web archiver. It doesn't make much sense if your notes are littered with links that don't work anymore. It's like a wine collector opening his cellar after a couple years only to find empty bottles.
December 9, 2025 at 7:00 PM
Note-taking app ecosystems should run a collective decentralized web archiver. It doesn't make much sense if your notes are littered with links that don't work anymore. It's like a wine collector opening his cellar after a couple years only to find empty bottles.
I’m surprised that I’m not surprised by the findings. They all line up with what system eval practitioners like @HamelHusain and @sh_reya have been saying.
x.com/dair_ai/sta...
x.com/dair_ai/sta...
December 7, 2025 at 6:15 PM
I’m surprised that I’m not surprised by the findings. They all line up with what system eval practitioners like @HamelHusain and @sh_reya have been saying.
x.com/dair_ai/sta...
x.com/dair_ai/sta...
An articulation of a reason why software engineers get such diverse results from LLMs, and why some large swath of engineers still think it’s all hype.
x.com/intuitmachi...
x.com/intuitmachi...
December 7, 2025 at 3:55 PM
An articulation of a reason why software engineers get such diverse results from LLMs, and why some large swath of engineers still think it’s all hype.
x.com/intuitmachi...
x.com/intuitmachi...
Surprised to run into a CSS bug that not even GPT-5.1 high could solve. Should have tried it with Claude Opus.
December 7, 2025 at 2:07 AM
Surprised to run into a CSS bug that not even GPT-5.1 high could solve. Should have tried it with Claude Opus.