What a Week for Generative AI
📝 Editorial
We just experienced one of the most active weeks of the year in the AI market. AWS came out with a lot of interesting announcements at re:Invent, PyTorch 2.0 was released and the NeurIPS conference unveiled a lot of interesting research initiatives from the most prestigious AI labs in the world. And yet, the most interesting headlines of the week seem to have centered, once again, on the generative AI space.
The generative AI market is accelerating at a frantic pace but last week was impressive even by those standards. The most interesting release of the week came from OpenAI with the announcement of ChatGPT which is part of the initiatives around the GPT 3.5 model. ChatGPT is a language model optimized for dialogue which means that it can follow the context of a conversation at a very deep level, answer follow up questions and even admit its mistakes. Google was also active in the generative AI space this week with the release of DreamFusion, a text-to-3D model that can produce incredibly realistic results. Not surprisingly, generative AI was front and center of the NeurIPS conference with many papers and announcements on that topic. Among those, NVIDIA published Get3D, another generative 3D model that can produce complex textured 3D meshes. NVIDIA also unveiled MineCLIP, a foundation model that was pretrained in YouTube footages of Minecraft videos and was then used to train a reinforcement learning agent able to complete several tasks on Minecraft.
Generative AI and foundation models continues to make headlines and are rapidly becoming one of the most exciting areas of the AI space.
🔺🔻TheSequence Scope – our Sunday edition with the industry’s development overview – is free. To receive high-quality content about the most relevant developments in the ML world every Tuesday and Thursday, please subscribe to TheSequence Edge 🔺🔻
🗓 Next week in TheSequence Edge:
Edge#249: we continue our series about ML interpretability by discussing model-intrinsic vs. post-hoc methods. We discuss activation atlases, a visual interpretability method developed by Google and OpenAI and cover the famous TensorBoard toolset.
Edge#250: Let’s deep dive into CICERO, Meta AI’s latest super model that is capable of negotiating and mastered the famous Diplomacy game.
🔎 ML Research
ChatGPT
As part of the GPT 3.5 series, OpenAI unveiled the research around ChatGPT, a conversational model that is optimized to follow up instructions and engage in complex dialogues —> Read more.
DeepNash
DeepMind published a research paper discussing DeepNash, a model-free reinforcement learning model that mastered the complex game of Stratego —>Read more.
DreamFusion
Google Brain published a paper detailing DreamFusion, a text-to-3D generative model driven by diffusion methods —>Read more.
NVIDIA Generative AI Research
At the NeurIPS conference, NVIDIA presented different papers that showcased the power of generative AI —>Read more.
🤖 Cool AI Tech Releases
PyTorch 2.0
PyTorch reached its 2.0 version with major new capabilities and performance improvements —>Read more.
SageMaker New Features
At its re:Invent conference, AWS unveiled a lot of additions to the SageMaker platform. Among those, we have new support for geospatial data, a new UI for SageMaker Studio, better notebooks, model cards, support or shadow testing and improved AutoML capabilities —>Read more.
VQ-Diffusion
HuggingFace announced support for VQ-Diffusion, a conditional diffusion model that operates on the quantasized latent space for its noising and denoising operations —>Read more.
🛠 Real World ML
Stability AI and AWS
Stability AI doubled down on its investment in AWS infrastructure —>Read more.
PyTorch in Meta’s Mobile experiences
Meta AI shared some information about their use of PyTorch to power different mobile experiences —>Read more.
LinkedIn’s Skills Graph
LinkedIn discussed the ML methods used to build its popular skill graph feature —>Read more.
💸 Money in AI
Data protection startup Shield raised $20M to expand its AI capabilities to protect data leaks.
Andrew Ng’s incubated Netail raised $5 million to enhance its AI platform for retailers.
Speech intelligence startup Deepgram raised $47 million to complete their series B.
Image annotation platform V7 raised $33 million making it another well-funded startup in the data labeling space.
AI-based customer intelligence platform Frame AI raised $7. 6 million to accelerate its growth.
Data warehouse robotics company Locus Robotics raised a monumental $117 million to continue competing with Amazon in that hot market.