Qwen and ChatGPT both flub counting the e's in 'seventeen'
Express_Quail_1493 · reddit · 2026-10-31
A side-by-side screenshot shows both qwen3.8-27b running locally and ChatGPT failing to count the e's in "seventeen": the former answered "4" while misspelling the word letter by letter, the latter confidently said "3." The correct answer is 2. A classic reminder that character-level counting remains an LLM failure mode, for local small models and frontier models alike.
Related event: Qwen3 and GPT Both Fail to Count the e's in 'seventeen'(2 posts)→
More from Fun
- Altman's claim that personal PC RAM is wasteful draws 'greed' backlash — pixlpa · 2026-10-31
- VLC's VideoLAN blisteringly rebukes French lawmaker over digital tax criticism — ngxson · 2026-10-31
- Altman says personal RAM is wasteful, mocked with '640KB is enough' comparisons — michalmalewicz · 2026-10-31
- LLM Acknowledgments in Bio Preprints Hit 17% in Three Years, Near Specialized AI Tools — teortaxesTex · 2026-10-31
- Neuralink user explains why your inner monologue stays private — XFreeze · 2026-10-31
- Meme: which startup is building agentic tools to make AI shut up — eigenron · 2026-10-31