oroborositron.bsky.social
Machine learning, D&D, music
Currently at The Coalition studio, Vancouver.
27 posts
111 followers
338 following
Regular Contributor
Conversation Starter
comment in response to
post
⚠️Risk of clams!⚠️
comment in response to
post
Great article. I also want to add that this sort of research is also super helpful in industry, where we attempt the same mistakes over and over because no one published about them. Looking forward to your article part 2.
comment in response to
post
What a badass trailer! Can't wait to hear more. I've definitely wondered how this project would shape up eventually. All the best!
comment in response to
post
Simply beautiful. Thank you for your amazing work.
comment in response to
post
Perfection!
comment in response to
post
Can't wait! I just hope we can read it too. I must be one of the last people on earth who cannot sit through a podcast. What do you do with the rest of yourself while listening? How do you do other things while paying attention to the podcast? Mysteries of life I'll never understand...
comment in response to
post
However, you can find quantized versions of those models that can be run on good single retail GPUs. The Ollama tool (and many others) also lets you try out quantized versions of several great (more) open models without needing to run inference code.
comment in response to
post
Some positive counter examples working towards full transparency are AI2 (Allen Institute of AI), EleutherAI, etc.
For trying out powerful models HuggingFace is a good starting point. To find powerful open models check out a leaderboard like lmarena.ai. Most SOTA large models need GPUs to run. 2/?
comment in response to
post
There is a vibrant community of people working on such models. Challenge is hardware availability for training. So permissively licensed models are usually trained by corporations. They rarely share data sources/sets or training code+tricks. They often release model weights and inference code. 1/?
comment in response to
post
That was such a great talk (just watched it)! Was drawn to the mention of Viewpoints. Used it in grad school years ago to help analyze/structure motion improvisation in a human-AI "dance" installation piece. So cool to see that framework used in the games space. Learn something new every day! 😀
comment in response to
post
And to you and yours, Eric! 🥂
comment in response to
post
It's fantastic! I'm not familiar with the LoL world or characters, so I considered myself a relatively neutral observer going in. I really enjoyed it!
comment in response to
post
Good to see you on here!
comment in response to
post
Hey hey! Being upbeat and doing ML is pretty much my exact system prompt. 😅 Can you add me too, please?
comment in response to
post
This is really cool! Thanks for the intuitive explanation and visualisation.
comment in response to
post
Is it out already? Have you played it? Oooh! I've been under a rock it seems... Can't wait!