Deep Transformers without Shortcuts: Modifying Self-attention for Faithful Signal Propagation He et al.: https://
arxiv.org/abs/2302.10322 #Artificialintelligence #DeepLearning #Transformers
RESEARCH
-
Deep Transformers Without Shortcuts: Self-Attention Signal Propagation
By
–
-
Interpretability and Symbolic Components in AI Alignment
By
–
Personally, I think that interpretability may be essential for alignability and debuggability, but that pure neural networks—ie those without some symbolic components—are unlikely to ever give us that. We may have to give up some performance to achieve that.
-
The overwhelming daily volume of AI research professionals must absorb
By
–
The amount that people in AI have to read and stay up to date on (I don’t mean people interested in AI, I mean people dedicating their entire career to its development)… I just – My brain feels hot at the end of each day. That’s how much it is to take in. Every day.
-
Understanding Technology History as Continuous Knowledge Revolution
By
–
the proper way to understand the history of technology is not four separate technological revolutions but a single big continuous revolution enabled by the ability, and the desire, to figure things out and to compound that knowledge
-
ChatGPT as Benchmark: Language Precision for Conversational AI Systems
By
–
Good point! Super interesting to see ChatGPT become the new benchmark. It'll be challenging but important to find the right language moving forward when talking about new conversational systems – in a way that doesn't conflate those aspiring to be ChatGPT with ChatGPT equivalents
-
Oxford’s RealFusion Enables 360° Object Reconstruction from Single Images
By
–
Oxford U Presents RealFusion: 360° Reconstructions of Any Object from a Single Image
-
AI as a Medium: Future Capabilities Depend on Human Engineering
By
–
"What will {AI | computers} be able to do in the future?" Well, absolutely anything you can make them do. They're a medium, not a specific technique. And the capabilities of both are 100% dependent on the human engineers that make them do those things.
-
AI is too vague a term like computers
By
–
Talking about "AI" in the abstract is a bit like talking about "computers" — it could mean anything.
-
DeepMind Welcomes Jeff Clune to Research Team
By
–
Thrilled to welcome @jeffclune to @DeepMind – we've long admired Jeff's brilliant work, and we're very excited to finally be getting the chance to work together!