Joined April 2013
- Tweets 102,974
- Following 249
- Followers 301
- Likes 97,238
4,163 Photos and videos
My user interactions with gippity now result in every serious question i ask getting a response that starts, "I'm treating this as a (rephrased) x and not a (rephrased) y." Can't tell if that's common or if I've battered the poor thing into stating constraint space.
12
Jun 16
If you need a term at least I don't feel silly saying out loud, "rail hopper" comes from a long line of massacring one great German word into a weaker English counterpart.
Jun 16
The outrageous effectiveness of Leitwörter
I've realised that all of the great skills I've written share one thing in common.
They make heavy use of Leitwörter - leading words.
A leitwort comes from literary theory. It's a repeated word or phrase used throughout a text to establish a theme or anchor meaning.
In skills, a leitwort is a word or phrase the agent uses to guide its own behavior. In other words, it's a word that leads the agent in a certain direction.
Let's take the leitwort "zone of proximal development" from my /teach skill. It's a phrase from the study of education. It means the "zone where the user feels challenged but not overwhelmed".
I use this only a couple of times throughout the skill's SKILL.md, but I've seen it almost every time the agent invokes the skill.
- "Let me adjust the lesson so it's in the user's zone of proximal development."
- "I'll read the learning records to establish the user's zone of proximal development."
In other words, that single phrase encodes how the agent should behave, in a concise token the agent can itself repeat to reinforce its own behavior.
Not only that, but it also likely tickles the agents' parameters related to educational research and "being a good teacher".
For engineering, leading words like "tracer bullets", "deep modules", "test seams", "clean code" are outrageously effective for leading the agent to produce better code.
So a leitwort in AI is any word or phrase you use that appears in the agents' thinking traces and guides its behavior.
Enjoy finding your own.
16
Jun 14
All of @mattpocockuk ideas about skills as agentic steers comes from a cognitive reverse engineering of how they'll be applied that treats SKILL.md as an api surface with non-determinative results and a black box harness. Would be interesting to see an A/B of read/invoke.
1
46
Jun 14
When an agent treats a skill in invocation, it is a pseudcode script with defined gates after invocation . The run is what's targeted first . No pre-flight matters except as it limits the agent from starting the process . Once it is locked into SKILL process, it never leaves.
1
49
Jun 14
The goal state for a SKILL for an agent is not the branching path it or the user are used to: it is to treat the skill like a pseudocode script, steer on failure with the options presented and produce a result.
18
Jun 12
Anyway, never let your heroes post cross-field ontological claims when you respect theirs within the space they normally occupy. Otherwise, well...
May 18
You have no idea whether that's true and projected it onto an imaginary viewer as a tautology. For someone respected by his peers for acknowledging the value of craft and diligence within a discipline, you really enjoy making posts that would make Ryan Fleury's of other fields
32
Screenclutter retweeted
Jun 12
Is that a Goose in a Mexican football shirt at the world cup...🤣🤣🤣
129
1,431
16,476
4,580,300
I love NixOS. "Everything has to be determinative." No it doesn't, I have to add the right files via git add. Never says I have to put the git file or repo somewhere safe. Check mate, hoes.
2
31
"Sometimes the sirens are loud, and loud is bad. And the freedom is only this: you check who the signal was addressed to. If it wasn't you, you don't have to let it in. It was never made for you."
1
16
For what it's worth, that's how an LLM responded to an AGENTS.MD that didn't worry about ablation but just told them to keep jumping until the conversation tree felt stable in <2,000ks.
12
May 24
Is there a @GaryMarcus for sane people? Like an active observer of the AI space as a whole, rather than a use-case contrarian, who actually interacts with the space but doesn't require cable news hits because he no longer does research and is irrelevant otherwise?
6
13
5,225
May 24
He thinks he's @NickKristof and able to parachute into anythíng but it just isn't the same thing.
1
206
May 23
RT @OGBlackRedGuard: Dale Earnhardt was the most redneck redneck to ever redneck and he took the confederate flag off his car because his B…
5,868
May 23
I won't tag him in the call-out of a popular account whose app is basic electron slop for anti-openclaw snobs. but if you want mobile access to your agents, there isn't anythíng better than litter from @SIGKITTEN and it isn't even close.
1
8
1,401
May 23
optimizing remote control for "desktop users who don't use other cloud options and want to use another fully functional computer to handle the devload" is something OSS is perfect for: solving for niches unresponsive to how a majority of users actually would like it to work.
May 22
I made a mistake. I underestimated Julius again.
T3 Code's remote feature is so far ahead of the remote control options offered by, well, everything else.
2 clicks to get a URL. Now I'm running a bunch of worktrees on my Mac Mini. Tailscale support built in too.
1
83
May 23
I made the mistake of trying to set up T3code so I could check on a single repo from my phone. Good fucking luck.
20
May 17
The crucial thing about model-agnosticism is that you have to actually understand what a user might do within a similar environment. Haters and evangelists alike refuse to adopt that basic frame when assessing AI.
1
30