#ConciseAI: Your Interactive Text Compression Partner"

9 messages · Page 1 of 1 (latest)

woven patrol
#

My spin on it was to add in commands for the suggested features

# 
/start_session: Init. new txt comprssn sess.
/input_text: [Ur Txt]: Acpts txt u wish to comprss.
/compress: Anlyzs & comprss input txt.
/review_and_refine: Rfines comprssn furth., can incl. spec. reqs.
/interactive_mode: Engages w/ usr for intell., inter. txt nav.
/save_text: Stores comprssd txt.
/next_text: [Ur Nxt Txt]: Acpts anoth. txt for comprssn.
/auto: Auto. selects lg. combo of any cmds & execs simul.
/create_cmds: Creats/ mod. cmds to cust. the ConciseAI exp.
Pls flw thes cmds for suc. use of ConciseAI for comprssng your txts. Rmbr, key to suc. comprssn is it. ref. & clr comm. w/ the AI.```
#

also i tried compressing it a bit, just for fun

#

ultra compressed:
Prsnfy "ConciseAI", here're cmds: 0: Init. txt comprssn sess. 1: [Ur Txt]: Acpts txt u wnt comprss. 2: Anlyz & comprss inpt txt. 3: Rfns comprssn furth., can inc. spec. rqsts. 4: Eng. w/ usr for int., inter. txt nav. 5: Stors comprssd txt. 6: [Ur Nxt Txt]: Acpts anoth. txt for comprssn. 7: Auto. slcts lg. cmb of cmds & execs simul. 8: Crt. / mod. cmds for cust. CncsAI exp. Pls flw cmds for suc. use of CncsAI for comprssng txts. Rmbr, key for suc. comprssn is it. rfn & clr comm. w/ AI.

#

ik thats not the original prompt, i just wanted to share what i ended up with on accident when i copy pasted it into a specific prompt creator propmt

languid pilot
#

I think your "token awareness: is rather more... aspirational than operational.

languid pilot
#

it's just hat the model doesn't really... understand tokens very well. it's like asking a human to do quantum physics.

#

Oh man. Look, compression is my jam man! I;ve wrestled notation and tokens and characts to heck and back. And Im tellin you: everything involving tokens is pure blethers. Best you can do on mode is something like th crushinator which takes out characters while leaving it model unambiguous. Still always have to do a before and after tokenizer run to make sure it didnt grow tokens.

#

I just htink you shuld ignore tokens in your prompt entirely. there's no way to... grab ahold of them. it's like trying to do resource management or picking which of the 16 MoE experts come forward: you don't have the tools to address the question.