hedge@beehaw.org to Technology@beehaw.orgEnglish · 1 year agoSomething weird going on at DuckDuck…beehaw.orgimagemessage-square34fedilinkarrow-up192arrow-down10file-text
arrow-up192arrow-down1imageSomething weird going on at DuckDuck…beehaw.orghedge@beehaw.org to Technology@beehaw.orgEnglish · 1 year agomessage-square34fedilinkfile-text
minus-squareDonnieDarkmode@lemm.eelinkfedilinkEnglisharrow-up31·1 year agoYeah where are those descriptions coming from? Also mentions “the strike workers’ strike” and repeats “politics” twice
minus-squarefloofloof@lemmy.calinkfedilinkEnglisharrow-up44·edit-21 year agoSeems like a poorly trained ML model - crudely speaking, perhaps its training set was tilted towards descriptions of conservative news sites so it learned to insert the word “conservative” when describing a news site.
minus-squaretuckerm@supermeter.sociallinkfedilinkarrow-up30·1 year agoI don’t want to jump to conclusions, but that does sound like a very possible explanation. Poo. I was hoping DDG would keep LLM-generated summaries out of their UI.
minus-squarenyan@lemmy.cafelinkfedilinkEnglisharrow-up17·1 year agoMy guess is that they’re surfacing something from Bing rather than doing this themselves. Still Not Good, though.
minus-squareradix@lemm.eelinkfedilinkarrow-up10·1 year agoThat makes it sound like an outdated LLM, like GPT-2 or something. You think it is?
minus-squarehedge@beehaw.orgOPlinkfedilinkEnglisharrow-up4·1 year agoAlas, not smart enough to know 🙁
Yeah where are those descriptions coming from? Also mentions “the strike workers’ strike” and repeats “politics” twice
Seems like a poorly trained ML model - crudely speaking, perhaps its training set was tilted towards descriptions of conservative news sites so it learned to insert the word “conservative” when describing a news site.
I don’t want to jump to conclusions, but that does sound like a very possible explanation.
Poo. I was hoping DDG would keep LLM-generated summaries out of their UI.
My guess is that they’re surfacing something from Bing rather than doing this themselves. Still Not Good, though.
That makes it sound like an outdated LLM, like GPT-2 or something. You think it is?
Alas, not smart enough to know 🙁