SCOREDevtools & code
JevSpeak
Luigi@LuigiProof
0:25JevSpeak enables TypeSafe's Jev decision model to generate sentences by answering semantic questions and compiling those decisions into words using code, without any large language models.
TakeJevSpeak has it answer 13 semantic questions in parallel — intent, stance, claim, tone — each as a probability distribution, then a deterministic compiler turns those decisions into sentences.
I built a project that makes Jev talk.
Jev (TypeSafe's decision model) can't generate a single word. JevSpeak has it answer 13 semantic questions in parallel — intent, stance, claim, tone — each as a probability distribution, then a deterministic compiler turns those decisions into sentences.
No LLM anywhere in the loop.
Meaning by Jev. Words by code.
Its uncertainty shows in the wording: 0.54 → "maybe", 0.86 → "probably", below 0.5 it declines instead of bluffing. Switch to 中文 and Jev doesn't change at all.
@typesafeai @CompleteSkeptic
🔗 jevspeak.org
💻 github.com/MM-sheng/jevspeak