What's New In ML #5
Using ML to talk to animals, aligning AGI with humans, ML in space, making robots dance, text-to-video generators, Tesla's supercomputing architecture, Google's chatbot, ML for mineral exploration
Hello friends,
Welcome to the 5th edition of What’s New In ML on Infinite Curiosity. The goal of this segment is to help you stay updated on what’s happening in the world of ML, AI, and Data. We’ll talk about real-world applications, cutting edge research, startups, and everything else that’s moving this field forward.
If you have a question, submit it here and I’ll get back to you with my thoughts. Subscribe to this newsletter to receive it in your inbox every week:
APPLICATIONS AND RESEARCH
Machine Learning is helping scientists talk to animals
Scientists are using machine learning to eavesdrop on naked mole rats, fruit bats, crows and whales — and to communicate back. Scientists have begun deploying this technology to decode animal communication, using machine-learning algorithms to identify when squeaking mice are stressed or why fruit bats are shouting. Even more ambitious projects are underway — to create a comprehensive catalog of crow calls, map the syntax of sperm whales and even to build technologies that allow humans to talk back.
Looks like someone got inspired by Doctor Dolittle. This could be an interesting development if it's used in a good way. Read more
OpenAI proposes an approach to ensure AGI is aligned with humans
At a high-level, our approach to alignment research focuses on engineering a scalable training signal for very smart AI systems that is aligned with human intent. It has three main pillars:
- Training AI systems using human feedback
- Training AI systems to assist human evaluation
- Training AI systems to do alignment research
OpenAI has proposed this framework to ensure that AGI (Artificial General Intelligence) remains aligned with humans. And that we have a framework to keep track of it. Read more
CMU paper on orientation of galaxies
Galaxies in the Universe are sparsely scattered through space with no fixed pattern or regular geometry and do not fit with conventional approaches of fully connected layers, convolutional neural networks, or recurrent neural networks to represent the data as vectors, grids (tensors), or sequences (ordered sets), respectively. Additionally, we aim to capture the correlated alignments in a population of galaxies. Thus, graphs are a natural way to model galaxies in the Universe. Graphs are defined as sets of nodes and each of the links connecting pairs of nodes.
Researchers at CMU have trained a neural network to predict the coherent orientations of galaxies. Read more
Microsoft releases a library of pre-trained humanoid control models
Microsoft Research’s Robot Learning group is releasing MoCapAct, a large library of pre-trained humanoid control models along with enriched data for training new ones. This will enable advanced research on artificial humanoid control at a fraction of the compute resources currently required.
This research will help humanoid robots dance like regular people. Or even Mick Jagger. Read more
Text-to-video generators
The computation cost is exponentially higher for text-to-video generation, which makes the training from scratch nearly unaffordable. The lack of relevant datasets also adds to the problem. However, researchers across the globe are now slowly breaking these barriers.
Systems like DALL-E 2, Midjourney, and Stability Diffusion can generate images from text input. So naturally the next step is to generate videos from text input. Here are some active projects in this field: Runway's AI-assisted video editor, Deepmind's Transframer, Microsoft's NUWA Infinity, and CogVideo. Read more
Tesla is building its own supercomputing architecture for AI and data
"Right from my interview with Elon, he asked me what can you do that is different from CPUs and GPUs for AI. I feel that the whole team is still answering that question." This led to the development of the Dojo training tile, a self-contained compute cluster occupying a half-cubic foot capable of 556 TFLOPS of FP32 performance in a 15kW liquid-cooled package.
Dojo is Tesla's fully custom supercomputing architecture. It's basically a massive composable supercomputer. Read more
New neuromorphic chip to reduce energy consumption for AI on edge devices
The NeuRRAM chip is the first compute-in-memory chip to demonstrate a wide range of AI applications while using just a small percentage of the energy consumed by other platforms while maintaining equivalent accuracy.
AI systems are energy intensive. And energy is precious when we're dealing with edge devices. This can help. Read more
Google has opened up the waitlist to talk to its experimental AI chatbot
Earlier this year, Google unveiled AI Test Kitchen — an Android app that lets users talk to one of its most advanced AI chatbots, LaMDA 2. Today, the company is opening up registrations for early access.
Meta did it with BlenderBot 3. Now Google is doing it with AI Test Kitchen. Read more
ML could revolutionize mineral exploration
Using a global data set of zircon trace elements, new research demonstrates the power of machine learning algorithms to accurately identify and locate porphyry copper deposits.
Researchers have presented two ML techniques to identify deeply buried deposits. Read more
Paragraph AI launches the world's first AI writing assistant powered by GPT-3 in the App Store
ParagraphAI can generate essays, articles, emails, messages and more with perfect spelling, grammar, tone and vocabulary. Its release comes just in time for the back-to-school season with an NLP model and integration that generates 99% original plagiarism-free drafts. Still, in order to get the best results from ParagraphAI, they recommend that users incorporate their own edits and perform both plagiarism checks and fact checks once the text has been generated.
If a human has to write a long essay, creating the first draft can take hours. But AI-powered writing assistants can do it in a second. You can then spend your time editing it and making it your own. It's poised to change the way we think about writing. Read more
STARTUP FUNDING
Headroom raises $9M led by Equal Opportunity Ventures for its AI-powered videoconferencing platform
Qloo raises $15M Series B led by Eldridge and AXA for its AI platform for culture and taste preferences
Lumachain raises $19.5M Series A led by Bessemer for its AI platform for food supply chains
Inworld AI raises $50M Series A for its virtual character developer platform led by Intel Capital and Section 32
Zilliz raises $60M Series B extension led by Prosperity7 for its vector database
Anyscale raises $99M Series C funding co-led by Addition and Intel Capital for its distributed computing platform
If you are getting value from this newsletter, consider subscribing and sharing with your friends: