Generative AI record label, very nice.
@aibreakfast
-
Meta unveils ImageBind, a multi-modal model
By
–
Meta's new ImageBind model binds data from six modalities at once (images and video, audio, text, depth, thermal and inertial measurement units) without the need for explicit supervision.
— AI Breakfast (@AiBreakfast) 9 mai 2023
This is the precursor to creating your own metaverse. pic.twitter.com/YyVJjgyOPqMeta's new ImageBind model binds data from six modalities at once (images and video, audio, text, depth, thermal and inertial measurement units) without the need for explicit supervision. This is the precursor to creating your own metaverse.
-
Voice cloning and translation between Lex Fridman and Jordan Peterson
By
–
Demo:
— AI Breakfast (@AiBreakfast) 8 mai 2023
Voice cloning with language translation between @lexfridman and @jordanbpeterson pic.twitter.com/oc0KPWf8tNDemo: Voice cloning with language translation between @lexfridman and @jordanbpeterson
-
Google could lead the open source AI initiative
By
–
Ironically Google could actually lead this initiative – they know they don't have a competitive product with Bard, and they saw how fast the dev community pounced on LLAMA models once it was "leaked" from Meta. Google could gain a significant amount of public trust (and
-
Open-source LLMs must dominate to prevent control
By
–
There needs to be a massive push to make open-source LLMs the dominate product. A future where one or two companies essentially control the information the world receives has been done before, and it never ends well. Release the models.
-
Research: Startups’ Appeal Versus Industry Giants
By
–
Money may be plentiful, but the top researchers don't want to be on the losing team – when OpenAI started gaining traction, they jumped ship and went with the "scrappy startup".
-
Conspiracy theory about the AI race
By
–
AI Race Conspiracy Theory: The Google "We Have No Moat" leaked document (essentially admitting that they are losing the AI race) was not a leak. It was written with the full intention of being public – it was just disguised as a leak. Here's why: 1. It redirects attention:
-
OpenAI prepares Text-to-3D with SHAP-E
By
–
Yes. OpenAI releasing Text-to-3D soon: SHAP-E They dropped a research paper and Github code for a model called SHAP-E for creating 3D models through text prompts. We are probably 1-2 months away from seeing Text-to-3D printers, as in text-to-object.
-
Shap·E: OpenAI’s 3D generative model
By
–
Shap·E is a conditional generative model for 3D assets, and you can read the full paper here: https://arxiv.org/pdf/2305.02463.pdf … GitHub: https://github.com/openai/shap-e
-
Neural Appearance Model for Realistic Textures
By
–
The appearance model uses learned hierarchical textures interpreted by neural decoders with two graphics priors, allowing for accurate reconstruction of mesoscale effects and efficient importance sampling. Paper: https://research.nvidia.com/labs/rtr/neural_appearance_models/
… Hi-res video: https://research.nvidia.com/labs/rtr/neural_appearance_models/assets/nvidia_neural_materials_video-2023-05.mp4
…