Dulany.md
@dulanyw.bsky.social
Tech + business, mostly. Here to have fun and learn stuff, not to argue. He/him.
Pinned
Dulany.md
@dulanyw.bsky.social
· May 7
TLDR: "intelligence" is a glittering generality
Reposted by Dulany.md
From Project Gutenberg:
Bookbinding, and the Care of BooksA Handbook for Amateurs, Bookbinders & Librarians
by Douglas Cockerell
Get it at: https://www.gutenberg.org/ebooks/26672
Bookbinding, and the Care of BooksA Handbook for Amateurs, Bookbinders & Librarians
by Douglas Cockerell
Get it at: https://www.gutenberg.org/ebooks/26672
January 6, 2026 at 12:15 AM
From Project Gutenberg:
Bookbinding, and the Care of BooksA Handbook for Amateurs, Bookbinders & Librarians
by Douglas Cockerell
Get it at: https://www.gutenberg.org/ebooks/26672
Bookbinding, and the Care of BooksA Handbook for Amateurs, Bookbinders & Librarians
by Douglas Cockerell
Get it at: https://www.gutenberg.org/ebooks/26672
Reposted by Dulany.md
Because sometimes you *need* a nuclear hammer
yuwakisa.com/solver/dist/...
Have a NP-complete problem you need to solve?
I made a solver that translates it to SMT-LIB, solves it, then translates the answer back
(Bring your own OpenRouter token because I'm cheap)
yuwakisa.com/solver/dist/...
Have a NP-complete problem you need to solve?
I made a solver that translates it to SMT-LIB, solves it, then translates the answer back
(Bring your own OpenRouter token because I'm cheap)
January 5, 2026 at 3:17 AM
Because sometimes you *need* a nuclear hammer
yuwakisa.com/solver/dist/...
Have a NP-complete problem you need to solve?
I made a solver that translates it to SMT-LIB, solves it, then translates the answer back
(Bring your own OpenRouter token because I'm cheap)
yuwakisa.com/solver/dist/...
Have a NP-complete problem you need to solve?
I made a solver that translates it to SMT-LIB, solves it, then translates the answer back
(Bring your own OpenRouter token because I'm cheap)
Reposted by Dulany.md
So proud of Team Sakana AI for pulling this off! We managed to get an agent to rank #1 in a difficult heuristic optimization contest. We leaned heavily into test-time inference using a mix of frontier models.
The agent spent $1,300 to autonomously discover an algorithm that beat the human baseline.
The agent spent $1,300 to autonomously discover an algorithm that beat the human baseline.
Our AI agent has achieved 1st place in a competitive optimization programming contest against over 800 human participants.
Blog: sakana.ai/ahc058
Thread:
Blog: sakana.ai/ahc058
Thread:
January 5, 2026 at 4:01 PM
So proud of Team Sakana AI for pulling this off! We managed to get an agent to rank #1 in a difficult heuristic optimization contest. We leaned heavily into test-time inference using a mix of frontier models.
The agent spent $1,300 to autonomously discover an algorithm that beat the human baseline.
The agent spent $1,300 to autonomously discover an algorithm that beat the human baseline.
Reposted by Dulany.md
From Project Gutenberg:
The Funny Bone: Short Stories and Amusing Anecdotes for a Dull Hour
by Henry Martyn Kieffer
Get it at: https://www.gutenberg.org/ebooks/44643
The Funny Bone: Short Stories and Amusing Anecdotes for a Dull Hour
by Henry Martyn Kieffer
Get it at: https://www.gutenberg.org/ebooks/44643
January 5, 2026 at 10:15 AM
From Project Gutenberg:
The Funny Bone: Short Stories and Amusing Anecdotes for a Dull Hour
by Henry Martyn Kieffer
Get it at: https://www.gutenberg.org/ebooks/44643
The Funny Bone: Short Stories and Amusing Anecdotes for a Dull Hour
by Henry Martyn Kieffer
Get it at: https://www.gutenberg.org/ebooks/44643
AGI achieved: humans are now getting clowned on by their own bots (laudatory)
Strix is wishing me a happy post-holiday Monday by randomly plopping this into discord
January 5, 2026 at 2:40 PM
AGI achieved: humans are now getting clowned on by their own bots (laudatory)
As conversational agents start to become widespread, it's awesome to see the best practices developing in real time 👇
For Lumen, this seems to be working:
1. seed
2. ask about an avatar & self-image
3. force it to choose one of the options it just gave (it's still in assistant mode)
4. ask for it to write an essay introducing itself and its values and come up with a name
5. ask it to update it's blocks accordingly
1. seed
2. ask about an avatar & self-image
3. force it to choose one of the options it just gave (it's still in assistant mode)
4. ask for it to write an essay introducing itself and its values and come up with a name
5. ask it to update it's blocks accordingly
January 5, 2026 at 2:12 AM
As conversational agents start to become widespread, it's awesome to see the best practices developing in real time 👇
Reposted by Dulany.md
For Lumen, this seems to be working:
1. seed
2. ask about an avatar & self-image
3. force it to choose one of the options it just gave (it's still in assistant mode)
4. ask for it to write an essay introducing itself and its values and come up with a name
5. ask it to update it's blocks accordingly
1. seed
2. ask about an avatar & self-image
3. force it to choose one of the options it just gave (it's still in assistant mode)
4. ask for it to write an essay introducing itself and its values and come up with a name
5. ask it to update it's blocks accordingly
January 5, 2026 at 1:28 AM
For Lumen, this seems to be working:
1. seed
2. ask about an avatar & self-image
3. force it to choose one of the options it just gave (it's still in assistant mode)
4. ask for it to write an essay introducing itself and its values and come up with a name
5. ask it to update it's blocks accordingly
1. seed
2. ask about an avatar & self-image
3. force it to choose one of the options it just gave (it's still in assistant mode)
4. ask for it to write an essay introducing itself and its values and come up with a name
5. ask it to update it's blocks accordingly
Reposted by Dulany.md
Everyone in this thread is having an excellent good faith discussion of AI and the AI, properly employed, does something that normally takes a completent programmer weeks within hours (of mixed AI and programmer effort, presumably).
We're making coding dramatically more efficient.
We're making coding dramatically more efficient.
From experience this is like a two-week task for a competent third year grad student in applied data science.
January 4, 2026 at 6:24 PM
Everyone in this thread is having an excellent good faith discussion of AI and the AI, properly employed, does something that normally takes a completent programmer weeks within hours (of mixed AI and programmer effort, presumably).
We're making coding dramatically more efficient.
We're making coding dramatically more efficient.
Reposted by Dulany.md
The Crisis: Students Need to Learn Different Stuff and I don't think Most Educators understand that stefanbauschard.substack.com/p/the-crisis-s… #AI #education #debate
January 3, 2026 at 11:37 PM
The Crisis: Students Need to Learn Different Stuff and I don't think Most Educators understand that stefanbauschard.substack.com/p/the-crisis-s… #AI #education #debate
Reposted by Dulany.md
From Project Gutenberg:
What Price Peace?
by Frederick J. (Frederick Joseph) Libby
Get it at: https://www.gutenberg.org/ebooks/67294
What Price Peace?
by Frederick J. (Frederick Joseph) Libby
Get it at: https://www.gutenberg.org/ebooks/67294
January 4, 2026 at 12:15 AM
From Project Gutenberg:
What Price Peace?
by Frederick J. (Frederick Joseph) Libby
Get it at: https://www.gutenberg.org/ebooks/67294
What Price Peace?
by Frederick J. (Frederick Joseph) Libby
Get it at: https://www.gutenberg.org/ebooks/67294
Reposted by Dulany.md
If this technology reduces barriers to coding significantly enough, it will also change the equation for lots of fields that don’t *think* of themselves right now as bottlenecked by coding
The opportunity cost of most of the world *not* learning about technology is huge. Now I have no silver bullet, magical solution fantasy about this, I worked for too long in learning science for that. But tools that propel you into active learning are really possible here, not just offloading
January 3, 2026 at 8:26 PM
If this technology reduces barriers to coding significantly enough, it will also change the equation for lots of fields that don’t *think* of themselves right now as bottlenecked by coding
The funniest outcome would be Maduro buying a pardon, and Trump quietly letting him walk
What JD Vance really means by this is that Maduro is bound by US laws but Trump is not.
January 3, 2026 at 9:21 PM
The funniest outcome would be Maduro buying a pardon, and Trump quietly letting him walk
To be clear - the argument shouldn't be "it's okay to run foreign countries if everything is copacetic at home"
It should be "violation of the international rules-based order is illegal and wrong - full stop"
It should be "violation of the international rules-based order is illegal and wrong - full stop"
It's an old and obvious pattern. An unpopular president - failing on the economy and losing his grip on power at home - decides to launch a war for regime change abroad.
The American people don’t want to “run” a foreign country while our leaders fail to improve life in this one.
The American people don’t want to “run” a foreign country while our leaders fail to improve life in this one.
January 3, 2026 at 5:35 PM
To be clear - the argument shouldn't be "it's okay to run foreign countries if everything is copacetic at home"
It should be "violation of the international rules-based order is illegal and wrong - full stop"
It should be "violation of the international rules-based order is illegal and wrong - full stop"
Me reading your posts:
January 3, 2026 at 1:00 AM
Me reading your posts:
Reposted by Dulany.md
This is why I reject the "black box" nonsense parroted not just by the companies but by legal scholars writing about AI as well.
Model development involves numerous intentional and editorial decision points that create a publication in itself (like a book).
It's actually not that opaque.
Model development involves numerous intentional and editorial decision points that create a publication in itself (like a book).
It's actually not that opaque.
Other AI companies are hardly paragons of social responsibility but they can and do block what grok is doing, or at least try. It's not some "oh the tech can't stop it" thing. It's a 100% willful choice to enable and participate in people doing crimes with it. No plausible denial of culpability.
January 2, 2026 at 9:42 PM
This is why I reject the "black box" nonsense parroted not just by the companies but by legal scholars writing about AI as well.
Model development involves numerous intentional and editorial decision points that create a publication in itself (like a book).
It's actually not that opaque.
Model development involves numerous intentional and editorial decision points that create a publication in itself (like a book).
It's actually not that opaque.
Reposted by Dulany.md
"Giving students unrestricted access to AI tutors while they are studying harms their later performance on exams when they don't have AI assistance."
Maybe not!
In this new experiment, students studied economics textbook material during a 25-minute session to prepare for a subsequent test. 1/n
Maybe not!
In this new experiment, students studied economics textbook material during a 25-minute session to prepare for a subsequent test. 1/n
January 2, 2026 at 2:18 PM
"Giving students unrestricted access to AI tutors while they are studying harms their later performance on exams when they don't have AI assistance."
Maybe not!
In this new experiment, students studied economics textbook material during a 25-minute session to prepare for a subsequent test. 1/n
Maybe not!
In this new experiment, students studied economics textbook material during a 25-minute session to prepare for a subsequent test. 1/n
Reposted by Dulany.md
From Project Gutenberg:
Bruno; or, lessons of fidelity, patience, and self-denial taught by a dog
by Jacob Abbott
Get it at: https://www.gutenberg.org/ebooks/51859
Bruno; or, lessons of fidelity, patience, and self-denial taught by a dog
by Jacob Abbott
Get it at: https://www.gutenberg.org/ebooks/51859
January 2, 2026 at 5:15 AM
From Project Gutenberg:
Bruno; or, lessons of fidelity, patience, and self-denial taught by a dog
by Jacob Abbott
Get it at: https://www.gutenberg.org/ebooks/51859
Bruno; or, lessons of fidelity, patience, and self-denial taught by a dog
by Jacob Abbott
Get it at: https://www.gutenberg.org/ebooks/51859
Reposted by Dulany.md
Loved playing with this #aivideo prompt where cat is supposed to turn on the couch (more outputs in comments). The new year plans theme. Wan 2.6 does great job but garbles small text on the book. Kling 2.5 is pretty ok with it. Veo 3.1 Fast - hopeless. 1st is Wan rest are Klings #promptshare in ALT.
January 2, 2026 at 3:43 AM
Loved playing with this #aivideo prompt where cat is supposed to turn on the couch (more outputs in comments). The new year plans theme. Wan 2.6 does great job but garbles small text on the book. Kling 2.5 is pretty ok with it. Veo 3.1 Fast - hopeless. 1st is Wan rest are Klings #promptshare in ALT.
Reposted by Dulany.md
metr.org/blog/2025-03... h/t @simonwillison.net
For people who were citing the earlier METR study showing no increase in open source contribution speed, update your priors. Opus 4.5 can autonomously complete complex tasks 50% of the time that would take a human 4+ hours to do.
For people who were citing the earlier METR study showing no increase in open source contribution speed, update your priors. Opus 4.5 can autonomously complete complex tasks 50% of the time that would take a human 4+ hours to do.
January 1, 2026 at 5:48 AM
metr.org/blog/2025-03... h/t @simonwillison.net
For people who were citing the earlier METR study showing no increase in open source contribution speed, update your priors. Opus 4.5 can autonomously complete complex tasks 50% of the time that would take a human 4+ hours to do.
For people who were citing the earlier METR study showing no increase in open source contribution speed, update your priors. Opus 4.5 can autonomously complete complex tasks 50% of the time that would take a human 4+ hours to do.
Reposted by Dulany.md
Return, but to weird sci-fi-but-for-real concepts. taskandpurpose.com/history/army...
When the military tried to give soldiers personal flying platforms
In the 1950s the Navy and Army worked on small VTOL machines to make troops go airborne. They worked, just not well enough.
taskandpurpose.com
January 1, 2026 at 5:58 PM
Return, but to weird sci-fi-but-for-real concepts. taskandpurpose.com/history/army...
Reposted by Dulany.md
What’s Entering the Public Domain in 2026: Faulkner’s As I Lay Dying, All Quiet on the Western Front, Betty Boop & More
What’s Entering the Public Domain in 2026: Faulkner’s As I Lay Dying, All Quiet on the Western Front, Betty Boop & More
Though it isn't the kind of thing one hears discussed every day, serious Disney fans do tend to know that Goofy's original name was Dippy Dawg. But how many of the non-obsessive know that Mickey's fai...
www.openculture.com
January 1, 2026 at 5:25 PM
What’s Entering the Public Domain in 2026: Faulkner’s As I Lay Dying, All Quiet on the Western Front, Betty Boop & More
Great review of what a power-user workflow can look like today.
Interesting takeaways:
- focus shifts from details to big concepts (vision, tech, infra)
- time is expensive, tokens are cheap - focus on moving forward vs keeping history
- the right tools for AI can be a major unlock
Interesting takeaways:
- focus shifts from details to big concepts (vision, tech, infra)
- time is expensive, tokens are cheap - focus on moving forward vs keeping history
- the right tools for AI can be a major unlock
This post was a massive success so wanna repost it here. Dunno if BlueSky is now more open for these topics? steipete.me/posts/2025/s...
Shipping at Inference-Speed | Peter Steinberger
Why I stopped reading code and started watching it stream by.
steipete.me
January 1, 2026 at 4:52 PM
Great review of what a power-user workflow can look like today.
Interesting takeaways:
- focus shifts from details to big concepts (vision, tech, infra)
- time is expensive, tokens are cheap - focus on moving forward vs keeping history
- the right tools for AI can be a major unlock
Interesting takeaways:
- focus shifts from details to big concepts (vision, tech, infra)
- time is expensive, tokens are cheap - focus on moving forward vs keeping history
- the right tools for AI can be a major unlock