Pavel AI
@pavelai.bsky.social did:plc:fbjzu4nxfapyyzpwtfody3az
Teach and write about AI and development
Bluesky’a katılma tarihi:
- Takipçi
- 55
- Takip edilen
- 60
- Post
- 413
- Sıralamaya girdiği gün
- 0
Özet
Pavel AI (@pavelai.bsky.social), 24 Kasım 2024 tarihinde katılan bir Bluesky kullanıcısı. 55 takipçisi var, 60 hesabı takip ediyor ve 413 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.
This is weird and surprising to me. It seems unlikely models were directly trained or RL'd to do this, yet it seems like it's able to do it--which seem aligned with the claim that neural networks are already clean and interpretable. x.com/rikiyatakeh...
Postu göster> We give GPT-6 Astra the weights, a description of the scene and one recorded episode [...] In a single try with no tools, Astra’s edit makes the car succeed in all 1,024 episodes, changing 3.5% of the [173K!!] weights. rikiyat.github.io/blog/weight-space-p...
Postu göster> In Weight-Space Programming, a general-purpose LLM reads and writes a network’s weights directly (i.e., not through an optimizer, and in particular not with gradient descent) around some natural language spec, sometimes even with no environment feedback at all. rikiyat.github.io/blog/weight-...
Postu gösterLess than 1 tok/s but still cool to see people innovating here.
Postu gösterDeepInfer-Mac showcases running a 750B-class AI model on a MacBook Air with just 24 GB of memory, using expert routing and caching techniques to generate text at impressive speeds despite storage limitations. https://arxiv.org/abs/2610.36564
Postu göster
Blokladığı hesaplar
Bluesky’da bloklar herkese açık kayıtlardır.