I make shadurs with a 'u', and I also do ML stuff. https://nitter.gq/t.co/F1iLxFROjf https://nitter.gq/t.co/HZcpV1omv4
In a Treehouse
Joined June 2020
- Tweets705
- Following288
- Followers943
- Likes7.2K
Pinned Tweet
I haven't posted anything in a while...
So here's an implementation of PointPillars, a 3D object detection network using point clouds all done in #VRChat! It's similar to the stuff you see on self driving cars.
Code: github.com/SCRN-VRC/Point-Pi…
#unity3d #MachineLearning
SCRN 🏳️⚧️ retweeted
Remember this meme from 2022 which contrasted DALL·E “mini” against DALL·E 2?
We’re going to see the same thing now for large language model: Community-developed LLMs, some without any forced ‘alignment’ or ‘moralizing’ censorship, vs closed mainstream LLMs run by big companies.
ALT "AI Image models are very powerful, so we'll be rolling out DALL·E 2 to a select group of artists at first... We will monitor for no people, violence to ensure a safe and..." vs "Minions at cross burning", "Jar jar Binks at the Nuremberg trials", "Jeffrey epstein funko pop", "gender reveal 9/11"
SCRN 🏳️⚧️ retweeted
Scaling Transformer to 1M tokens and beyond with RMT
Recurrent Memory Transformer retains information across up to 2 million tokens.
During inference, the model effectively utilized memory for up to 4,096 segments with a total length of 2,048,000 tokens—significantly exceeding the largest input size reported for transformer models (64K tokens for CoLT5 (Ainslie et al., 2023), and 32K tokens for GPT-4 (OpenAI, 2023)). This augmentation maintains the base model’s memory size at 3.6 GB in our experiments
abs: arxiv.org/abs/2304.11062
github: github.com/booydar/t5-experi…
I had fun taking a break from shadurs to make a quick trailer for a friend.
Go check out the world, it looks amazing!
#VRChat
Omg whats this? a release announcement?!
I am happy to say my new VRChat world Oia Lookout is now live!
vrch.at/b2vt268q
#VRChat #VRChatWorld
1
6
596
SCRN 🏳️⚧️ retweeted
I am releasing version 2.0.0 of my optimizer.
Now VPM & UPM compatible!
Additionally it merges simple toggles in the fx layer into one big direct blend tree.
There is also improved merging of blendshapes that get animated in a fixed ratio to each other.
SCRN 🏳️⚧️ retweeted
How neat. Researchers claim to have found, at long last, an "einstein" tile - a shape that tiles the plane in a pattern that never repeats. First two authors appear to have no academic affiliations. arxiv.org/abs/2303.10798
SCRN 🏳️⚧️ retweeted
Presenting Long Suffering by @ElIZAGRACEMUSIC This is one of my favorite songs. I loved making this performance with the help of @TheHoizen and cinematography by @Depereo_
This song is a strong emotional one. I hope I conveyed those feelings through this
youtube.com/watch?v=-lXEY52i…
SCRN 🏳️⚧️ retweeted
Announcing GPT-4, a large multimodal model, with our best-ever results on capabilities and alignment: openai.com/product/gpt-4
SCRN 🏳️⚧️ retweeted
Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models
build a system called Visual ChatGPT, incorporating different Visual Foundation Models, to enable the user to interact with ChatGPT by 1) sending and receiving not only languages but also images 2) providing complex visual questions or visual editing instructions that require the collaboration of multiple AI models with multi-steps. 3) providing feedback and asking for corrected results
abs: arxiv.org/abs/2303.04671
SCRN 🏳️⚧️ retweeted
At #magfest this year, we used an #esp32s2 on our Swadges and were able to play with people in #vrchat. youtube.com/watch?v=VnZac6lA…
SCRN 🏳️⚧️ retweeted
Today we release LLaMA, 4 foundation models ranging from 7B to 65B parameters.
LLaMA-13B outperforms OPT and GPT-3 175B on most benchmarks. LLaMA-65B is competitive with Chinchilla 70B and PaLM 540B.
The weights for all models are open and available at research.facebook.com/public…
1/n
