when contexts are long, attending to every single token in the past feels wasteful (and not at all how human brains work). feels like a natural setting for compression… DMC seems like a huge improvement in transformer inference speed — congrats to the authors!
TECHNOLOGY
-
Cerebras WSE-3: The Fastest AI Processor, Outperforming H100
By
–
Cerebras’ third-generation wafer-scale engine (WSE-3) is the fastest AI processor on Earth, and makes the H100 look like an abacus in comparison: 4 trillion transistors, 125 petaflops of computing power
-
Web designers and developers shape tomorrow’s digital future
By
–
"As designers and developers for the web (and beyond), we’re responsible for building the future every day, whether that may take the shape of personal websites, social media tools used by billions, or anything in between.
-
Shaping the Web’s Future Amid Constant Transformation
By
–
Ste Grainer of @alistapart asks how we can shape the future of the web when it's constantly remaking itself.
-
Interface AI Innovation Requires GPT-5 Level Intelligence
By
–
The interface is great. It needs GPT-5 class brains to be truly transformative, I think, but it is the first example of a truly different mode of working with AI to accomplish work tasks. A really cool start.
-
Claude 3 Haiku Now Supports 200k Token Context Window
By
–
We have longer context windows for all Claude 3 models, including the just-released Haiku: https://
poe.com/Claude-3-Haiku
-200k
… -
Technology and Advocacy: Reshaping Innovation Through Social Impact
By
–
Looking forward to this! Technology as a tool for advocacy, and advocacy as a tool to (re)shape technology https://
x.com/OsloFF/status/
/OsloFF/status/1768363132903190702
… -
Collaborative Generative AI for 3D Modeling at GTC24
By
–
Take a deep dive into the collaborative process of developing a cutting-edge #generativeAI solution for 3D modeling at #GTC24. Discover the technology and challenges involved, lessons learned, and glimpse into the future with @Shutterstock
. https://
nvda.ws/43kTGIU -
LangChain 0.2 Release: Production-Ready Libraries Update
By
–
RFC: Expedited 0.2 release For the last ~6 months we've been pushing hard to make the LangChain libraries production ready. A big part of this has been splitting up packages to make them lightweight and easier to properly version. Since `langchain` (the package) 0.1 we've
-
AI-Powered Local Search Integration with Yelp Maps
By
–
We've rolled out improvements to local searches through our integration with @Yelp and maps to help you quickly find information on local restaurants, and businesses! pic.twitter.com/5WVc20yh8E
— Perplexity (@perplexity_ai) 14 mars 2024We've rolled out improvements to local searches through our integration with @Yelp and maps to help you quickly find information on local restaurants, and businesses!