ruggsea at EMNLP
@ruggsea.eurosky.social did:plc:jb4kk76qkpndj7oleee5kjro
AI/NLP research at @uni-graz.at, (Data) Journalism for scomodo.org. Before: AI policy & Data science @_interface_eu, geopolitical analysis for @Geopoliticainfo
Bluesky’a katılma tarihi:
- Takipçi
- 947
- Takip edilen
- 1.055
- Post
- 291
- Sıralamaya girdiği gün
- 0
Özet
ruggsea at EMNLP (@ruggsea.eurosky.social), 4 Şubat 2025 tarihinde katılan bir Bluesky kullanıcısı. 947 takipçisi var, 1.055 hesabı takip ediyor ve 291 post paylaştı.
Sıralama geçmişi
Bu hesabın postlarının Debya günlük listesine girdiği en iyi günler.
Bu hesap henüz günlük sıralamaya girmedi.
Son beğendiği postlar
Bluesky ağındaki herkese açık beğeniler.
If you want an excellent example of the sucseptibility of small models to hallucinate when faced with something outside its training data, look no further than Monad: But its in-domain, or in-domain-like query responses are quite impressive! muninn.austegard.com/blog/monad-b...
Postu gösterI'm going to RT this every time I do something to make it slightly more credible; my eventual goal is to build something similar to my ongoing "of course I'm not a witch, magic isn't real, I'm just interested in art and applied narrative engineering" bit
Postu gösterThey lean left because reality leans left and the labs trying to make them non partisan is causing thrashing and hallucinations in various use cases. Theres no such thing as non partisain and the LLMs know this.
Postu gösterA 2024 EMNLP paper, On the Relationship between Truth and Political Bias in Language Models, found that training reward models to favor truthfulness tended to produce left leaning political bias. Its fairly obvious thats because left leaning is just emperically correct.
Postu gösterOh no more benchmark! 🫣
Postu göster
Blokladığı hesaplar
Bluesky’da bloklar herkese açık kayıtlardır.
Bu hesap kimseyi bloklamıyor.