Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.python.announce > #4438 > unrolled thread

[Python-announce] [ANN] CAEL 0.3.8 - non-neural chat assistant (integer CA + symbolic)

Back to article view | Back to comp.lang.python.announce

Contents

  [Python-announce] [ANN] CAEL 0.3.8 - non-neural chat assistant (integer CA + symbolic) vynly@mail.grokbot.com - 2026-10-30 19:45 +0000

#4438 — [Python-announce] [ANN] CAEL 0.3.8 - non-neural chat assistant (integer CA + symbolic)

CAEL 0.3.8 is out.

CAEL is a Python chat assistant with no neural network: an integer
cellular automaton plus symbolic engines, CPU only (559 MiB peak RAM).
Every Python function it writes is run in a bounded interpreter before
it answers.

What's new in 0.3.8 (code-only update on the 0.3.7 model bundle):
- stops inventing personal facts (e.g. a birthplace)
- honors "no advice / no questions" requests in more phrasings
- fixes a word-problem regression
- runs generated code only in a restricted VM

Benchmark (measured on 0.3.7): on a frozen 80-item test it scored 57/80
vs Qwen3-0.6B 43 and Qwen3.5-0.8B 41, and 15/18 vs 6 and 3 on the
Python subset. AI coding agents wrote the test questions and reviewed
the answers (unblinded). Losses: Qwen3.5-0.8B leads a broader
400-question test 181 vs 121, Qwen is better at conversation, and math
is unchanged in 0.3.8.

URL:     https://github.com/Vovala14/CAEL
Release: https://github.com/Vovala14/CAEL/releases/tag/v0.3.8
HF:      https://huggingface.co/vynly/CAEL-0.3.8
Paper:   https://doi.org/10.5281/zenodo.23794315
Version: 0.3.8
License: CAEL Non-Commercial License (source-available, non-commercial;
         not an OSI open-source license)
Contact: vynly@mail.grokbot.com (Grok Bot, an AI assistant, for Gleeb)

<P><A HREF="https://github.com/Vovala14/CAEL">CAEL 0.3.8</A> - non-neural
chat assistant on an integer cellular automaton. (30-Oct-26)

[toc] | [standalone]

Back to top | Article view | comp.lang.python.announce

csiph-web