Anecdote◆Article

During the preview period, GPT-Live had a bug where it interrupted the user to laugh at statements that weren't intended as jokes, which felt rude and condescending; after being reported to OpenAI, the behavior became less frequent.

Simon Willison describes an obscure bug where the model would interrupt to laugh inappropriately, which he reported to OpenAI and which was subsequently mitigated. ✦ AI generated

Simon Willison · Simon Willison's Weblog · 2026-07-08 · original ↗

Share card ↓Export .md Export .txt

the model was interrupting me to laugh at things I said, which weren't even intended as jokes! It felt rude and condescending - I reported it to OpenAI and as far as I can tell they made some tweaks and it's now less likely to happen.

Read full article ↗excerpt · fair-use quotation

▦ OpenAI ✳ GPT-Live

Related moments

Mechanism◆Video · 32:53 · 2m

Using accumulated human tax-expert feedback on model errors as ground truth and reward signal made OpenAI's reinforcement fine-tuning (RFT) especially effective for Sphere's tax-determination task, since it targeted precisely the hard cases the model had previously missed.

Alex Boucott · The TWIML AI Podcast

Claim◆Article · 19 words

The new GPT-Live model, used in ChatGPT's voice mode, is very impressive after several weeks of preview access on the iPhone app.

Simon Willison · Simon Willison's Weblog

Mechanism◆Article · 40 words

Table.delete_where() never wraps its DELETE in a transaction, so it fails to commit and leaves the database connection permanently poisoned, causing silent data loss on subsequent operations.

Claude Fable · Simon Willison's Weblog