Significant progress in AI and Robotics this week.
— Brett Adcock (@adcock_brett) June 1, 2025
So, I summarized everything from Google DeepMind, Anthropic, Figure, DeepSeek, Sakana AI, Perplexity, ETH Zurich, Unitree, Physical Intelligence, and more.
Here's everything you need to know and how to make sense out of it:
Google DeepMind announced new Gemma models, including:
— Brett Adcock (@adcock_brett) June 1, 2025
—SignGemma to convert sign language into spoken words, coming later this year
—MedGemma, a 4B param multimodal model and a 27B text-only model for medical text and image comprehensionpic.twitter.com/6b9Jc6huJx
Anthropic launched voice mode for Claude Sonnet 4
— Brett Adcock (@adcock_brett) June 1, 2025
Users will have 5 voice personalities to choose from, with an integration with Google Workspace
Currently in mobile beta for English-speaking userspic.twitter.com/ikour7Ab8D
Figure Update: Our next-generation humanoid, F.03, is officially walking
— Brett Adcock (@adcock_brett) June 1, 2025
We've also merged three separate teams into our AI group, Helix, to speed up how quickly our robots learn and scale into the markethttps://t.co/XhSl6YNogY pic.twitter.com/6Px30b1hUv
DeepSeek pushed an update to R1, instantly taking it to the #3 spot on the Artificial Analysis leaderboard
— Brett Adcock (@adcock_brett) June 1, 2025
Other notable model improvements include enhanced front-end capabilities, reduced hallucinations, and support for JSON output & function callinghttps://t.co/6y8QxLhHei pic.twitter.com/r50B86floS
Japan's Sakana AI dropped Darwin Gödel Machine, a self-improving agent that can modify its code to boost performance
— Brett Adcock (@adcock_brett) June 1, 2025
On SWE-bench, DGM improved its performance from 20.0% to 50.0%, while on Polyglot, it increased its success rate from 14.2% to 30.7%https://t.co/Y7IZVjTQZa pic.twitter.com/axNrZZvt7M
Perplexity launched Labs, a new tool for building interactive apps
— Brett Adcock (@adcock_brett) June 1, 2025
It uses Deep Research with tools like image generation to create everything from analytical reports to websites
Only available to Perplexity Pro users on iOS, Android, and the webpic.twitter.com/OkTETINouO
ETH Zurich just demonstrated a robot dog playing badminton using only onboard perception
— Brett Adcock (@adcock_brett) June 1, 2025
In the video, a single RL policy coordinates 18 DOF simultaneously, achieving 10 consecutive rally shots with 12.06 m/s swing velocity and sub-400ms reaction timepic.twitter.com/KWbiRXNyzP
Physical AI announced knowledge insulation, a way to train vision-language action models 7.5x faster with diffusion output
— Brett Adcock (@adcock_brett) June 1, 2025
This enables the model to inherit better language following from the VLM, leading to better resultspic.twitter.com/X76h6xzJwB
Hume announced EVI 3, a speech-language model that can understand and generate any human voice and personality from a prompt in <1s
— Brett Adcock (@adcock_brett) June 1, 2025
It uses a voice-to-voice architecture and comes with a deeper understanding of tune, rhythm, timbre, and speaking stylepic.twitter.com/eMVPcGZEMY
Robotics developer Igor Kulakov just demoed MicroFactory, a $5K boxed frame with two AI-powered robotic arms
— Brett Adcock (@adcock_brett) June 1, 2025
These arms use visual demos and modular grippers to perform repetitive assembly tasks, right from screwing nuts to solderingpic.twitter.com/y84hFzuZ74
Resemble AI dropped Chatterbox, a SOTA open-source voice cloning model that includes:
— Brett Adcock (@adcock_brett) June 1, 2025
—Text-to-speech & voice conversion
—Emotion exaggeration control
—Ultra-low latency for real-time apps
—Imperceptible neural watermarkingpic.twitter.com/FaeujJmTN9
Black Forest Labs dropped FLUX.1 Kontext, an AI that generates both text and images as input, enabling:
— Brett Adcock (@adcock_brett) June 1, 2025
—In-context generation and 8x faster editing
—Character preservation, local editing, style transfer, and maintaining consistency across image versionshttps://t.co/Q2fciLNdkV pic.twitter.com/4cCadvlj9m
The Robot Studio and Hugging Face unveiled HopeJr, an ultra-affordable open-source humanoid
— Brett Adcock (@adcock_brett) June 1, 2025
Costing just $3K, HopeJr can walk and manipulate many objects with 66 actuated degrees of freedom
Full bill of materials and links to source parts on GitHubpic.twitter.com/0F3KmSgSz7
We're hiring for hundreds of roles @Figure_robot:
— Brett Adcock (@adcock_brett) June 1, 2025
> AI Engineers (many)
> Staff Security Engineer
> HMI Design Lead
> System Integration & Test (many)
> Legal (many)
> Manufacturing (many)
Apply here: https://t.co/CmraocEEAIhttps://t.co/vXj0wfA2dg
Welcome to Shanghai. https://t.co/3mvDF1Djjm
— Ambassador Chen Song (@PRCAmbNepal) June 2, 2025
A far cry from universal Mao jackets.
— Paramendra Kumar Bhagat (@paramendra) June 2, 2025
2: DeepTechMaxxing (Rohan Pandey) https://t.co/8J7bpsqD6Q
— Paramendra Kumar Bhagat (@paramendra) June 2, 2025
After the conclusion of the Bucharest Nine and Nordic countries summit, I spoke with journalists and shared some details of today’s negotiations with the Russians in Istanbul: an unconditional ceasefire, the exchange of prisoners, the return of children, and, importantly, the…
— Volodymyr Zelenskyy / Володимир Зеленський (@ZelenskyyUa) June 2, 2025
AOC Soars In Popularity Nationally, But Constituents Ask: “Where’s Our Rock Star?”
The ‘Three-Punch Combo’ Behind Ukraine’s Spectacular Drone Strike on Russia
"Trump flipped on us": MAGA reacts to potential national citizen database
No comments:
Post a Comment