Gosplan Cosplay 
Taggart 
"If you want to improve the model's output, you can write skill files with more specific instructions!"
"Oh wow so like a file that tells the computer to do exactly what you want?"
"Yep!"
"You're never gonna believe this."
fuzzyfuzzyfungusThe imperium is actually super-against AI; but this seems like one of those "a small mind is easily filled with faith" situations they tell me about.
crispy branzino ☭ (sold out forever)@mttaggart make no mistakes
Rui Batista@mttaggart @vascorsd Ah fuck.. I thought those md files were generated by llms but just realized people actually write them?!!
D. Moonfire@ragb @mttaggart @vascorsd Clearly they use AI to write the agent.md and skill files based on an AI written set of instructions on how to write instructions which is an AI generated file to write AI generated instructions....
@dmoonfire @mttaggart @vascorsd The meaning of anything and everything will be lost somewhere and ai will rule the world based upon dumb text with stuppid markup parcing rules. This is justice..
Nikhil 🐧@mttaggart come on the gambling factor is fun
ikeacurtains@mttaggart the sad fact is, the bosses prefer the left button.
labria@mttaggart @dtl skill.sh ;)
Curious CarrotIt's like people are desperately trying to extract deterministic behaviour from a stochastic rock.
Landa 
@curious_carrot And before that they desperately worked on turning perfectly fine deterministic rocks into stochastic rocks that pretend to chat.
@mttaggart
@mttaggart
smallbean@curious_carrot @mttaggart @curious_carrot @mttaggart weirdly enough i also think that they are trying to extract stochastic behavior (creativity, innovation, etc) from a deterministic rock (computer) at the same time
@mttaggart this brings to mind my reflection about retrieval augmented generation: “all you need to do to improve the knowledge of the model is give it a list of search results and have it paraphrase them”. Ok but what if I just got the results for myself. Is this not just handing a sheet of words to some guy so he can read them out?
Johnnyvibrant@mttaggart it really is like docker compose and bash scripts never existed. This is just the same...give the robot the commands it goes and does shit. If only people understood this and didnt just go running around like headless chickens saying its magic.
707Kat@johnnyvibrant I was thinking this exact thing.
"Remember when the program/compiler/shell just actually did what your file told it to? What a trip that was!"
@707Kat And if you did the same command twice...the same thing happened!!! What a trip
Raven Luni@mttaggart And to think I wasted my childhood writing Z80 opcodes on a bit of paper and entering them with POKEs ...
J. Alan Henning@RavenLuni Me, too! Which machine?
@jalanhenning 48k Speccy
Pseudo Nym
Ted Lemon
Mason Loring Bliss@RavenLuni @mttaggart I had a Z80 but I only had a book that taught 8080, so I did 8080 on it. I remember it fondly.
Anarchist Trans Ace Pirate"you know what would be cool? a large vehicle that multiple people get on and it drives around to the same stops every 15 minutes"
"you mean a bus?"
☮ ♥ ♬ 🧑💻@mttaggart all the work, none of the joy.
Alex o_O@mttaggart I will now refer to all files I create as skill files. 😎
@SpaceLifeForm I don't know what that means
It is about optimization.
You, being the coder, can provide a skill file telling the AI/LLM what it should do or not do. By providing the skill file, you are optimizing the model performance and trying to tell it what you want.
The same thing happens with real coding. You, being the coder, provide a source file to the compiler. The source file is a skill file that reflects the skill of your design and logic.
Your chances of getting the computer to do exactly what your want are much higher with the compiler.
(-o0 is a flag to the toolchain to compile for space saving at the machine code level as opposed to optimizing for speed. Sometimes, optimization for speed can result in subtle bugs. Rare)
Faraiwe
aburka 🫣
mahadevank@mttaggart that was 90% of all arguments I got when I told folks I don't use AI in my coding workflow directly.
Just write the specs
Have another AI review your PR
And they just don't seem to get the point that each AI step is error-prone and cannot be trusted.
LR@mttaggart writing a skill file to generate skill files 🤡
Phil@mttaggart Not to mention, you can put the skillfiles that you often use together for later re-use. Like a bookshelf or a room full of bookshelves. They should find a name for that.
shellsharks 🌙@mttaggart @mr_daemon Bringing all new layers to "skill issue" 😏
Eric LawtonIf the makers of AI™ find out you can do this, they might bias the output in their favour.
Luckily they are too ethical to do this.
</sarcasm>
Tom@mttaggart I was reading a pro-AI blogpost where the writer ended up submitting a 3200 word prompt, which resulted in a checklist for some activity which could easily be achieved another way. Absolute madness.