Latest update
No weights, no training: 1,000 AI agents wrote a 640,000-line if-else chatbot that ties a 360M-parameter model
First seen on X 27 hours ago@ANAg2bGOD ♥ 2,23025 views
Hugging Face has turned the old joke that "AI is just a bunch of if statements" into a real chatbot. PyBot runs on nothing but Python conditionals, regex patterns and lookup tables. A Japanese X post explaining it passed 2,200 likes.
According to the GitHub repo released Nov 16, a swarm of more than 1,000 AI agents wrote the code. They produced and evaluated over 5,000 versions, lifting PyBot's win rate on SmolTalkBench, where an LLM judge compares each reply with a reference answer, from 0% to 26%. Hugging Face says that puts it on par with SmolLM-360M-Instruct, a few years ago the state of the art under 1B parameters. Project lead Leandro von Werra said on X that 1,000 DeepSeek Flash agents wrote the whole thing in a week.
The result is about 640,000 lines of code plus 550,000 lines of tests the agents wrote themselves: 317,000 lines of world knowledge, 109,000 of conversation memory, 104,000 for composing replies. It needs only Python 3.8+ (python -m pybot), and the Hugging Face Space shows which module produced each answer. Just loading the code takes a few seconds.
Von Werra admits multi-turn chats and typos trip it up: "don't expect AGI, we only claim it's as good as a 360M model." One reply asked whether porting it to C would make a super-fast AI; another noted that when an answer is wrong, picking which of 640,000 lines to fix is still a human job. Northeastern's Arjun Guha shared a similar rule-based natural-language-to-code program that solves 16.6% of MBPP.
Comments 0
Select a sentence in the story (long-press on phones) to comment on it.
“For real Hugging Face actually went and physically implemented that old joke about AI being nothing but a bunch of if statements, using 1,000 agents.”
