diff --git a/archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on/2026-08-06-yacineMTB-i-think-the-ability-to-make-a-computer-program.md b/archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on/2026-08-06-yacineMTB-i-think-the-ability-to-make-a-computer-program.md new file mode 100644 index 0000000..2baf165 --- /dev/null +++ b/archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on/2026-08-06-yacineMTB-i-think-the-ability-to-make-a-computer-program.md @@ -0,0 +1,83 @@ +--- +title: "@VictorTaelin I think the ability to make a computer program progressively smaller (yet still output the same data) is close to the definition of intelligence, if not the definition itself" +type: archive +source: twitter +source_url: "https://x.com/yacineMTB/status/2085418818050912320" +author: "kache" +handle: yacineMTB +post_id: "2085418818050912320" +date: 2026-08-06 +archived: 2026-08-25 +draft: false +tags: + - archive + - twitter + - VictorTaelin +description: "@VictorTaelin I think the ability to make a computer program progressively smaller (yet still output the same data) is close to the definition of intelligence, if not the definition itself" +in_reply_to: "" +parent_post_id: "2085409601034371371" +--- + +## Source + +- URL: https://x.com/yacineMTB/status/2085418818050912320 +- Author: kache (@yacineMTB) +- Posted: 2026-08-06 17:32:46 + +## Branch + +**1/** **@yacineMTB** ^2085418818050912320 + +**@VictorTaelin** + +I think the ability to make a computer program progressively smaller (yet still output the same data) is close to the definition of intelligence, if not the definition itself + +**2/** **@VictorTaelin** ^2085421071168151730 + +**@yacineMTB** + +I agree 100% and I think AGI can be achieved by just training on that. The reason for my intuition is that I now concede LLMs can learn any algorithm they're exposed to, but we never exposed them to a task that demands all components of general intelligence. What you wrote does! + +**3/** **@yacineMTB** ^2085421434617098303 + +**@VictorTaelin** + +if only we could have some kind of verifier for this that we know won't halt.. + +**4/** **@VictorTaelin** ^2085421638166745477 + +**@yacineMTB** + +ohh now that sounds like a good idea + +give me one week! + +**5/** **@VictorTaelin** ^2085421797009248410 + +**@yacineMTB** + +also we need a single word for "fail to halt" + +**6/** **@slitdrip** ^2085434880389947678 + +**@VictorTaelin** **@yacineMTB** + +how about diverge? + +https://en.wikipedia.org/wiki/Divergence_%28computer_science%29 + +**7/** **@VictorTaelin** ^2085435867779195147 + +**@slitdrip** **@yacineMTB** + +oh well that works 🤦‍♂️ + +**8/** **@solodeprodigal** ^2085445707452891405 + +**@VictorTaelin** **@slitdrip** **@yacineMTB** + +💯 + +## Related + +- Spine: [[archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on]] diff --git a/archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on/index.md b/archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on/index.md new file mode 100644 index 0000000..83640da --- /dev/null +++ b/archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on/index.md @@ -0,0 +1,44 @@ +--- +title: "just saving my last conversation with Fable on this entire flattener debacle." +type: archive +source: twitter +source_url: "https://x.com/VictorTaelin/status/2085409601034371371" +author: "Taelin" +handle: VictorTaelin +post_id: "2085409601034371371" +date: 2026-08-06 +archived: 2026-08-25 +draft: false +tags: + - archive + - twitter + - VictorTaelin +description: "just saving my last conversation with Fable on this entire flattener debacle." +in_reply_to: "" +--- + +## Source + +- URL: https://x.com/VictorTaelin/status/2085409601034371371 +- Author: Taelin (@VictorTaelin) +- Posted: 2026-08-06 16:56:08 + +## Thread + +**1/** **@VictorTaelin** ^2085409601034371371 + +just saving my last conversation with Fable on this entire flattener debacle. I like it because it seems to be on the exact edge of what AI's still can't do (today) + +the task was to implement an "elegant" pattern-match flattener for Bend's core. the one Fable had written was extremely contrived (like, 3x longer than it should) and I was trying to get it to find the simplest possible implementation. + +note this problem is not new. converting pattern-matches into decision trees has been described in 1985. it is implemented in nearly every proof assistant. Fable knows how to build one. + +yet, what I demanded it something *slightly* different. a minimal implementation on Bend's context, which has a few nuances, like being based on point-free λ-trees. these nuances were enough to put it out of its zone of comfort, and, despite being an extraordinary model, it still couldn't make the jump from the textbook algorithm it knows, to a general solution on Bend's context + +that's why its code was so big. it was stuffing layers of duct taping, to compensate for the fact it couldn't see the general solution. 5 days, 15+ attempts, and the function barely got smaller. I even gave it a step-by-step derivation, but it STILL failed. ultimately, I just decided to write the algorithm myself. to its credit, my own solutions still had some gaps. I guess we're both equally stupid nowadays, but at least we managed to fix it together. below is this last conversation (summarized) + +this is a bit embarrassing because I burned millions of tokens and wasted a lot of time on something I honestly could do in an hour. that's stupid, and a good argument that AI is (still) a tool that requires skill, and can be misused in dumb ways that tank your productivity and end up being detrimental rather than good + +![](https://pbs.twimg.com/media/HPDc5vLWoAAXkOH?format=jpg&name=orig) + +Branches: [[archive/threads/VictorTaelin/2026-08-06-just-saving-my-last-conversation-with-fable-on/2026-08-06-yacineMTB-i-think-the-ability-to-make-a-computer-program]]