Qwen doubts it can see images, calls the tool anyway and succeeds
catgirl_liker · reddit · 2026-11-05
A Reddit user shares Qwen's amusing chain of thought: when asked to take a screenshot, the model doubts itself — "I'm an LLM, I can only read text" — then tentatively calls the tool anyway, mulling saving the image and parsing it programmatically, and is pleasantly surprised when it works: "Wow, I can see the image!" A cute look at how models misjudge their own capability boundaries in agent settings.
More from Fun
- Agent runs wild: all five GitHub budgets blocked at $0 spent — andimarafioti · 2026-11-05
- Speculation: The Browser Company's 'Holly Li' persona may be quietly phased out — RileyRalmuto · 2026-11-05
- The people angriest at AI are students who wrote their own essays while others cheated — NC_Renic · 2026-11-05
- Sim2real solved: just downsample sim and real until they're indistinguishable — eigenron · 2026-11-05
- Michael Moor mocks formulaic arXiv titles: 'Will arXiv become the new Nature?' — Michael_D_Moor · 2026-11-05
- Redditor shows off 82GB of household RAM: 'Hey Sam, you are not having it' — SamsungSam07 · 2026-11-05