Similar responses responses to the same prompt can be seen from o1-mini and Claude 3.5 Sonnet. Other models left as an engagement-bait exercise for the reader.
RESEARCH
-
GPT-4o tokenization splits ‘strawberry’ into single-character tokens
By
–
In both the prompt and response above, each letter in “strawberry” is assigned by the GPT-4o tokenizer to a separate single-character token, instead of being combined into subword-length tokens like “st” “raw” “berry” etc.:
-
Open Buildings Dataset Now Tracks Building Presence Over Time
By
–
An exciting update to our Open Buildings dataset. It now features building presence over time (as well as building height information). It does this by combining high resolution (50cm) but infrequent satellite images and low resolution (1000cm) but frequent (every 5 days). It… https://t.co/cswnXJIjAX pic.twitter.com/DjIxjZYYR5
— Jeff Dean (@JeffDean) 19 septembre 2024An exciting update to our Open Buildings dataset. It now features building presence over time (as well as building height information). It does this by combining high resolution (50cm) but infrequent satellite images and low resolution (1000cm) but frequent (every 5 days). It
-
Compiler Optimizations and Memory Management for FP16 Models
By
–
compiler optimizations, better memory management etc. same FP16 weights.
-
Anil Seth on Consciousness, Perception and Intelligence Research
By
–
I had the fortune of listening to
@anilkseth
today at Microsoft London on consciousness, perception, intelligence, life and diversity of human minds. Fascinating. Highly recommend his works. I’m glad that there are smart people like him exploring these research topics. -
Data versus Engineering: Large Robot Action Models Lecture
By
–
Thanks Stone! Glad you mentioned “good old fashioned engineering”. Have you seen this? Lecture: Is Data All You Need? Large Robot Action Models and Good Old Fashioned Engineering https://
bit.ly/Is_Data_All_Yo
u_Need
… -
OpenAI o1 Models Outperform Experts; Microsoft Launches Copilot Agents
By
–
Big News: OpenAI’s new o1 models (aka ‘Project Strawberry’) outperform experts on PhD-level problems. Microsoft boosted 365 apps with Copilot Agents & launched open-source benchmarks for AI. Explore these & more! #AI #Innovation
-
Video Generation AI Models: Benchmarks and Training Data Comparison
By
–
Great video generation. I wish we had good public benchmarks to compare to other approaches like Veo, Runway Gen-3, Sora, Haiper, etc. It would also be nice to know what data is being used for (pre-)training in each case, and to know how influential data and model scale are.… https://t.co/xbIXAk9q2L
— Nando de Freitas (@NandoDF) 19 septembre 2024Great video generation. I wish we had good public benchmarks to compare to other approaches like Veo, Runway Gen-3, Sora, Haiper, etc. It would also be nice to know what data is being used for (pre-)training in each case, and to know how influential data and model scale are.
-
First Agent Leaderboard Published with Multiple SEAL Rankings
By
–
This is the first of many Agent Leaderboards we plan to publish. Please follow along and see our other SEAL Leaderboards: