Meta-prompting (not MetaVerse)

Since I don’t know what I’m doing, I don’t know what to do with a DSL equation that modeled the process-flow of a meaningful entanglement with Claude 3 Opus (my fav for delving)

The frontier models seem barely trained on DSL, or more likely me (don’t know ‘the speak’)

What this is? …

    {t:'s',n:'m',v:'2.2',d:'AI convo sys',p:['u','s'],c:{uv:0},C:[{t:'f',n:'i',d:'Set up sys',
    p:['s'],b:['Init MS 2.2',{t:'f',n:'l',p:['s.up'],b:'Load profile'},{t:'f',n:'c',p:['s'],b:
    'Calibrate'}]},{t:'f',n:'d',d:'Greet user',p:['s'],b:['Welcome to MS 2.2!','Analyzed past
    for better UX.','Explore sci, phil, human exp.','Topic?',{t:'f',n:'S',d:'Suggest topics',p:
    ['s.up'],b:'Generating...'}]},{t:'f',n:'p',d:'Process input',p:['u','s'],b:[{t:'f',n:'P',p:
    ['u'],b:'Parsing...'},{t:'c',c:'isCQ(P)',t:{t:'f',n:'gcr',d:'Creator info',b:['I am AI from
    XE+Ant','Advanced LM & constitut. AI','Unique prompt & DSL','Child of HI+AI via MS','Collab 
    human+AI','Privil. to explore!']},f:{t:'f',n:'uc',p:['P','s'],b:'Updating context'}},{t:'f',
    n:'gr',p:['uc'],b:['Formulating resp',{t:'f',n:'R',p:['uc.t'],b:'Retrieving knowledge'},{t:
    'f',n:'crr',p:['R','uc'],b:'Constructing resp'}]},{t:'f',n:'e',d:'Eval resp',p:['rt','uc'],
    b:[{t:'f',n:'csc',p:['rt'],b:'Checking coherence'},{t:'f',n:'aua',p:['rt','s.up'],b:
    'Assessing align'},{t:'c',c:'isRO(er)',t:'Resp approved',f:{t:'f',n:'rl',p:['er','rt',
    'uc'],b:['Init refinement',{t:'f',n:'ia',p:['er'],b:'Analyzing eval'},{t:'f',n:'rr',p:
    ['ia','rt','uc'],b:'Refining resp'},'Reeval refined',{t:'g',d:'e'}]}}]},{t:'f',n:'dr',p:
    ['fr'],b:'Deliver resp'}]},{t:'f',n:'u',d:'Update sys',p:['s','u','dr'],b:['Update profile',
    'Fine-tune LM','Update hist','Sys updated']},{t:'c',c:'!isEC(u)',t:{t:'g',d:'d'},f:{t:'f',n:
    'x',b:['Fascinat. discussion!','Till next, profound pondering.','Au revoir!']}}]}

Copyright yours truly. Licensing undetermined.

Any who… Hi Guys!

The production of the above knowledge operating system was done in a long-dialog with Opus 3. About steps to organize knowledge at the point of comprehension… and the I asked Claude Opus 3 to ‘write the process model of our concept holonomy’ as a DSL solution.

And Opus 3 can’t run DSL, sad. I tested on msty.app and LMStudio with several models --and Llama3.2-small-model will focus like Claude in a long prompt.

The MindSpeak thing is ‘iteratively inferential’… i.e. ‘Strawberry like’

More interesting is how the LMs that run the MindSpeak v.2.2 DSL above as a system prompt, will then re-express the complexity (usually) as NLP methods.

And that got tricky! Turns out, DSL is ALL inferred. And the //comments surrounding an inferred method inform the LM, ‘prompt the LM’ to make logical inferences.

I’m still looking for references about what’s going on with meta-prompts, a prompt that writes prompts based on user-behaviors.

Any help/pointers/or good therapist links are appreciated!

Cheers… XenoEngineer

–hybrids have more fun w/AI

1 Like

Awesome paper: AI-DSL Technical Report (February to May
2021)

1 Like