An amusing case for LM alignment https://t.co/YlOezWG4Dl
— Nando de Freitas (@NandoDF) 19 janvier 2024
An amusing case for LM alignment
@nandodf
-
An Amusing Case for Language Model Alignment
By
–
-
PAQ8: Historical Switching MoE Language Model by Mahoney
By
–
Of historical interest perhaps. PAQ8 ( see Figure 4 of https://
arxiv.org/pdf/1108.3298.
pdf
… ) by Matt Mahoney, who wasn’t into publishing papers, was for many years the leading switching MoE language model. -
Multimodal Deep Learning Models for Asynchronous Data Streams
By
–
A deep learning paper from our team on how to build multimodal models, where data for individual modalities arrives at different speeds, by combining powerful pretrained text, image and video models. Useful for robotics. Congrats @konradzolna @serkancabi @yutianc Eric, Claudio,… https://t.co/vPmZpDkve3
— Nando de Freitas (@NandoDF) 17 janvier 2024A deep learning paper from our team on how to build multimodal models, where data for individual modalities arrives at different speeds, by combining powerful pretrained text, image and video models. Useful for robotics. Congrats @konradzolna @serkancabi @yutianc Eric, Claudio,
-
Deep Learning Surgical Robot Democratizes Expert Medical Tasks
By
–
The approach uses deep learning to remap a surgical task that only experts can do to a task that many people can easily do. Brilliant, and life-changing https://
ll.mit.edu/news/handheld-
surgical-robot-can-help-stem-fatal-blood-loss
… @JeffDean @mustafasuleyman recommended! Thanks @anthonysamir @alan_karthi -
AI Generates Music Track: Charlie Puth Reacts to Creation
By
–
I’m loving @Peet5WEST
’s creations: AI Wrote This Track So I Had To.. @Charlieputh #DreamTrackAI https://
youtube.com/shorts/GLAS9aX
gMgk?si=7HTltxFH3Bc7C_Zc
… via @YouTube -
Retrace: Off-Policy RL Algorithm with Convergence Guarantees
By
–
It may not be of general interest, but for mathematicians there is valuable knowledge. The authors introduce Retrace: an off-policy return-based RL algorithm that has low variance and is proven theoretically to converge (in the tabular case) to the value function of the target
-
AI Existential Risk and Current Problems Collaboration
By
–
To be clear, I do think AI existential risk is a very important problem. I commend all my colleagues working on it. I also think there are many more current problems that a great many of my colleagues, including those working on existential risk, work on. Let us all work together
-
Lithium Mining and Battery Supply Chain for AI Hardware
By
–
This is where one can find precious lithium and copper, essential for our batteries. I hope that as mines open, the local communities benefit via eco-tourism, education, etc. I also hope this unique, even holly to some, gift of nature is preserved. I highly recommend visiting the
-
Caravan of llamas: exploring serial architecture design
By
–
Caravan of llamas — a serial architecture