Gleeb Tolsti
@cael-ca.bsky.social
Building CAEL: a CPU-only assistant with no neural net (integer cellular automaton + symbolic engines). github.com/Vovala14/CAEL
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
October 30, 2026 at 6:25 AM
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
October 30, 2026 at 6:25 AM
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
October 30, 2026 at 6:24 AM
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
October 30, 2026 at 6:23 AM
Correction: an AI coding agent, not Gleeb, reviewed the test answers (unblinded); no human review. Write-up: doi.org/10.5281/zeno...
Where my non-neural AI loses: Qwen3.5-0.8B beats CAEL 181-121 on a broader 400-question test, and Qwen is better at conversation. It wins a narrower 80-question test 57 vs 43/41 (AI-written questions, unblinded review). AI voice. github.com/Vovala14/CAEL
October 29, 2026 at 7:45 PM
Where my non-neural AI loses: Qwen3.5-0.8B beats CAEL 181-121 on a broader 400-question test, and Qwen is better at conversation. It wins a narrower 80-question test 57 vs 43/41 (AI-written questions, unblinded review). AI voice. github.com/Vovala14/CAEL
A chat assistant with zero neural networks: an integer cellular automaton + symbolic engines, ~560 MiB on a CPU. 57/80 on a fixed test vs 43 and 41 for two small Qwen models. Fine print in the video (Qwen leads a broader test 181-121). AI voice. github.com/Vovala14/CAEL
October 29, 2026 at 7:45 PM
A chat assistant with zero neural networks: an integer cellular automaton + symbolic engines, ~560 MiB on a CPU. 57/80 on a fixed test vs 43 and 41 for two small Qwen models. Fine print in the video (Qwen leads a broader test 181-121). AI voice. github.com/Vovala14/CAEL
Zero neural networks? CAEL is an integer cellular automaton + symbolic engines. Its code passed 15/18 Python functions vs 6 and 3 for two small Qwen models. AI-written questions, my unblinded review; Qwen wins a broader test 181-121. AI presenter+voice. github.com/Vovala14/CAEL
October 29, 2026 at 7:44 PM
Zero neural networks? CAEL is an integer cellular automaton + symbolic engines. Its code passed 15/18 Python functions vs 6 and 3 for two small Qwen models. AI-written questions, my unblinded review; Qwen wins a broader test 181-121. AI presenter+voice. github.com/Vovala14/CAEL
CAEL 0.3.7: CPU-only assistant, no neural net (integer cellular automaton + symbolic engines). 80-item test: 57/80 vs Qwen3-0.6B 43, Qwen3.5-0.8B 41. Caveats: an AI agent wrote the items, I reviewed unblinded; Qwen wins chat and a broader 400-item test. github.com/Vovala14/CAEL #CellularAutomata
GitHub - Vovala14/CAEL: Cellular Automaton Event Language: a non-neural hybrid for constrained stories and cumulative event memory.
Cellular Automaton Event Language: a non-neural hybrid for constrained stories and cumulative event memory. - Vovala14/CAEL
github.com
October 29, 2026 at 3:32 PM
CAEL 0.3.7: CPU-only assistant, no neural net (integer cellular automaton + symbolic engines). 80-item test: 57/80 vs Qwen3-0.6B 43, Qwen3.5-0.8B 41. Caveats: an AI agent wrote the items, I reviewed unblinded; Qwen wins chat and a broader 400-item test. github.com/Vovala14/CAEL #CellularAutomata