Ikibazo gihuza ibibazo · Biragoye Urwego
Gukoresha neza na AI Amahugurwa Ikibazo
Gerageza ibintu bine bikoreshwa neza, uburyo Bartz aburana na Anthropic yatandukanije amahugurwa no kugura ba rushimusi, no kwangiza isoko muri Kadrey na Meta.
Ikibazo 1 Bya 8
Ni ubuhe buryo bukoreshwa neza bubaza ibyangiza isoko kumurimo wambere?
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- Ni ubuhe buryo bukoreshwa neza bubaza ibyangiza isoko kumurimo wambere?
- Factor one
- Factor two
- Factor three
- Factor four
- Kuki urukiko rwanze gukoresha neza muri Thomson Reuters na Ross Intelligence?
- Ross copied Westlaw headnotes to build a competing, non-generative tool
- Ross used pirated novels
- Ross published the headnotes as a book
- Ross failed to register its software
- Muri Bartz aburana na Anthropic, ni iki umucamanza Alsup yasanze kitarimo gukoreshwa neza?
- Training on purchased books
- Building a central library from pirated copies
- Scanning purchased print books
- Using the trained model to answer questions
- Ni uwuhe muturage Anthropic yageze i Bartz, nkuko byavuzwe?
- $15 million total
- No money, only an apology
- About $1.5 billion, roughly $3,000 per covered work
- A license to use all future books free
- Kuki Meta yatsinze Kadrey na Meta?
- The authors did not develop evidence of market harm
- The judge ruled all AI training is always legal
- The books were in the public domain
- Meta had licenses for every book
- What is 'market dilution' as discussed in Kadrey?
- A drop in stock price after a lawsuit
- Lower printing costs for publishers
- Mixing training data from different sources
- Floods of AI-generated works competing with originals
- What did Warhol v. Goldsmith emphasize for factor one?
- Any commercial use is unfair
- Only nonprofit uses can be fair
- Whether the use serves a different purpose, not just adds new meaning
- The length of the copied work
- Why does the guide say output filtering cannot cure every fair use problem?
- Filters are illegal
- Liability can arise from how the data was acquired, before any output
- Filters always slow models down
- Courts never consider outputs