Forget counting the Rs in strawberry, biggest challenge to LLMs is not making up bullshit about recent events not in their training data
antifuchs
The whisper model has always been pretty crappy at these things: I use a speech to text system as an assistive input method when my RSI gets bad and it has support for whisper (because that supports more languages than the developer could train on their own infrastructure/time) since maybe 2022 or so: every time someone tries to use it, they run into hallucinated inputs in pauses - even with very good silence detection and noise filtering.
This is just not a use case of interest to the people making whisper, imagine that.
This incredible banger of a bug against whisper, the OpenAI speech to text engine:
Complete silence is always hallucinated as "ترجمة نانسي قنقر" in Arabic which translates as "Translation by Nancy Qunqar"
Here’s Dave Barry, still-alive humorist, sneering at Google AI summaries, one of the most embarrassing features Google ever shipped.
Ooooh that would explain a similarly weird interaction I had on a ticket-selling website, buying a streaming ticket to a live show for the German retro game discussion podcast Stay Forever: they translated the title of the event as “Bleib für immer am Leben”, guess they named it “Stay Forever Live”? No way to know for sure, of course.
If you wanted a vision of the future of autocomplete, imagine a computer failing at predicting what you’re gonna write but absolutely burning through kilowatts trying to, forever.
https://unstable.systems/@sop/114898566686215926