🎉 Celebrity Clones, AI Crispr, Small Language Models, Talking Heads, Yet Another Robot
🎉 Celebrity Clones, AI Crispr, Small Language Models, Talking Heads, Yet Another Robot
Talent Agrees to Scanning for $$$, AI Gene Editing, New Powerfully Small Language Models, Deepfakes Galore, Boston Dynamics Robot is Flexible
Welcome to the third edition of AImpulse, a five point summary of the most significant advancements in the world of Artificial Intelligence.
Here’s the pulse on this week’s top stories:
What’s Happening: Leading Hollywood talent agency CAA has reportedly been testing an initiative called CAA Vault, allowing A-list clients to create AI clones of themselves to open new creative opportunities.
The details:
CAA partnered with AI firms to scan clients’ bodies, faces, and voices, creating AI replicas for uses like reshooting, dubbing, and stunt double superimposing.
CAA's goal is to eventually make the tech available industry-wide, not just to its clients.
Hollywood has already been bracing for AI’s impact, with Tyler Perry even halting studio expansions after seeing OpenAI’s Sora video capabilities.
Why it matters: CAA is taking proactive measures to help clients benefit from the shift to AI-first productions (inevitable). As models continue to improve, the difference between hiring Ryan Gosling or his AI replica may become imperceptible to the average fan.
What’s Happening: Profluent just developed OpenCRISPR-1, the world's first open-source AI-developed gene editor capable of editing the human genome.
The details:
Profluent trained LLMs on a vast dataset of diverse CRISPR systems to generate millions of new CRISPR-like proteins not found in nature.
OpenCRISPR-1 worked as well as or better than a naturally occurring editor.
CEO Ali Madani said the success points to a future where AI “precisely designs what is needed to create a range of bespoke cures for disease.”
Profluent open-sourced the model to make the gene-editing tech more accessible to researchers working on disease treatments.
Why it matters: This is an exciting space to watch — Gene editing already has the potential to revolutionize the medical field, and adding AI to the mix could take it to a whole new level. Tools like OpenCRISPR can eventually speed up innovation, reduce costs, and improve access to life-altering treatments.
What’s Happening: Microsoft just announced Phi-3, a new family of small language models that outperform rivals of larger sizes — reaching new benchmark milestones in the class of smaller-sized AI models.
The details:
The Phi-3 model family comes in three sizes: Phi-3-mini at 3.8B parameters, Phi-3-small at 7B parameters, and Phi-3-medium at 14B parameters.
Phi-3-mini’s benchmarks rival Mixtral and GPT 3.5 despite being significantly smaller, while also having a staggering 128k context window.
Mini’s 3.8B parameter size also enables the model to be deployed on-device while still maintaining quality and cost-effectiveness.
7B Phi-3-small and 14B Phi-3-medium are still in training and will also be available in the coming weeks.
Why it matters: The Phi-3-mini capabilities, in particular, are a massive breakthrough — opening the door for high-performing models to run locally on our phones.
What’s Happening: Microsoft just unveiled VASA-1, a new AI model capable of generating strikingly lifelike talking head videos from a single still image and an audio clip.
The details:
VASA-1 requires just a single photo and a speech audio file to create a realistic talking video of the subject, with synced mouth movements and emotive animations.
The model can generate nuanced expressions, natural head motions, and even realistic singing performances, going beyond simple lip-syncing.
Users can use input sliders to control aspects of the generated video, such as eye gaze direction, head distance, and emotional tone.
Why it matters: VASA-1 represents a big leap, with potential applications in areas like customer support, virtual avatars, gaming, and animation.
What’s Happening: Boston Dynamics just unveiled a sleek (or terrifying depending on who you ask), all-electric version of its Atlas humanoid robot, designed for real-world applications and boasting impressive new dexterity and agility.
The details:
The video showcases Atlas’ flexibility, including the ability to push itself up from a prone position and rotate its head and torso 180 degrees.
The release comes just one day after announcing the retirement of the original hydraulic Atlas robot.
The new Atlas features a more streamlined design than its predecessor, with a ring-light-like head and no more exposed cables.
Atlas will be tested in experiments with Hyundai (a Boston Dynamics investor) in the automaker’s factories.
Why it matters: With over a decade of experience in jaw-dropping demos, Boston Dynamics has done it again. Despite the ‘humanoid’ label, the new Atlas robot shows that the optimal design may not always be perfectly human — demonstrating mind-bending flexibility that strays away from human ability.