There was a lot of great parts in the original charter and announcement of OpenAI for sure I really liked this part on my side:
« Researchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code, and our patents (if any) will be shared with
@thom_wolf
-
OpenAI’s Original Charter: Open Research and Patent Sharing
By
–
-
OpenAI drama highlights need for decentralized AI development
By
–
Seeing the continuing OpenAI dramas over the past year I'm quite happy we don't depend on a single company to build AI responsibly for everyone
-
One Million AI Models Shared on HuggingFace Hub
By
–
One MILLION models shared openly on the HuggingFace hub And it’s growing by tens of thousands of models per week, of all size, all modalities, all specific domains, all licenses and origins. Seeing the open-source AI community strive like this is such a pleasure. Still
-
Cost-effective motors for multi-motor robotics development
By
–
One option would be to explore using even lower cost motors (like Feetech for instance) once a first version is working. With 13 motors in it, the price tag will never been extremely low though
-
Open-Duck-Mini Robot Achieves First Steps with Neural Networks
By
–
Open-duck-mini proudly standing up for the first time!
— Thomas Wolf (@Thom_Wolf) 16 septembre 2024
Neural network powered 💥
Join the discussion if you're interested in cute-RL and sim-to-real transfer for open-source robots (you can help even without robotic hardware)👇 https://t.co/JINvI4it5MOpen-duck-mini proudly standing up for the first time! Neural network powered Join the discussion if you're interested in cute-RL and sim-to-real transfer for open-source robots (you can help even without robotic hardware)
-
Mini BDX: Open-Source AI Weekend Project Released
By
–
cute weekend project with @antoinepirrone (fully open-source https://
github.com/apirrone/mini_
BDX
… – with some rough edges atm – tutorial will come) -
Anthropic Sonnet 3.5 Leap-frogs Competition in AI Models
By
–
still quite crazy how Anthropic leap-frogged everyone when they released Sonnet 3.5 a few months ago a model completely ahead of its time. we still have zero idea what special tricks they used to train it source: ARC-AGI evaluation of the new openai o1 https://
arcprize.org/blog/openai-o1
-results-arc-prize
… -
GPT-o1 Smaller Models Shift From Brute Force to Sequential Inference
By
–
GPT-o1 is also interesting in that it could keep pushing the trend of smaller (limited size) models, stopping the past trend to brute-force try to do everything system II in a system I single-forward-pass (very large models) and using sequential inference time compute instead
-
Video Language Models Unlock New Automation Possibilities
By
–
We've barely scratched the surface of what computer can automate when it comes to video
— Thomas Wolf (@Thom_Wolf) 12 septembre 2024
So many exciting use-cases for video+language models that are just outside of the realm of possible today: summarizing hours of videos, searching to find a specific moment/image/action/object… pic.twitter.com/wfdEFwMlhUWe've barely scratched the surface of what computer can automate when it comes to video So many exciting use-cases for video+language models that are just outside of the realm of possible today: summarizing hours of videos, searching to find a specific moment/image/action/object
-
SmolLM: Training Long Context Models at 0.5-1.5B Parameters
By
–
SmolLM for the win. Cool write up on training long context models around 0.5-1.5B parameters by the Jina team