This might be the best agent I've seen yet.
After raising $220M, @hcompany_ai just introduced an agent that can execute any task from a prompt.
Their "Runner H" can basically turn instructions into action with human-like precision.
Features:
โธ Navigates web interfaces with pixel-level precision.
โธ Interprets pixels and text to understand screens and elements.
โธ Automates workflows for web testing, onboarding, and e-commerce.
โธ Adapts automatically to UI changes.
โธ Achieves a 67% success rate on WebVoyager, outperforming competitors.
Architecture:
โธ Powered by a 2B-paramezer LLM for function calling and coding.
โธ Includes a 3B-parameter VLM for understanding graphical and text elements.
BREAKING . . . AND HUGE: A ruling just made by SDNY Judge Colleen McMahon dismissing @RawStory v @OpenAI (w/o prejudice) has enormous positive ramifications for #GenAI defendants in the District and potentially beyond. Here's why it may be seismic:
https://t.co/kBfmHg4uEJ
The former CEO of Sotheby's, Tim Smith. It is is job to know value. He retired and studied Bitcoin and the economy we have just been in.
Please I urge you to understand as Bitcoin goes from $76,000 to $1,000,000 and beyond.
Take 2 minutes to change your family future.
finally, AI trainings arrived to Krea.
train an AI model and teach it about your own characters, styles, products, and more.
sharing 100 early invites below ๐
We launched Voice Design to give creators limitless possibilities for creating unique character voices.
Here are some of our favorite creations so farโฆ
https://t.co/yqJ9zfcrP1
MASSIVE ๐คฏ
First dedicated transformer ASIC (Application-Specific Integrated Circuit) just dropped
Custom chip burns transformer architecture directly into silicon, making AI models run 10x faster than GPUs
What is Sohu ๐จโ๐ง ?
- Hardware architecture specifically optimized for transformer neural networks
- Processes AI models 10x faster and cheaper than GPUs
- Achieves >500,000 tokens/second throughput
โก Technical Capabilities
- Multicast speculative decoding
- Single core architecture
- Real-time content generation
- Parallel processing of hundreds of responses
- Beam search and MCTS (Monte Carlo Tree Search) decoding
- Supports MoE (Mixture of Experts) and transformer variants
๐ฏ Practical Applications
- Real-time voice agents processing thousands of words in milliseconds
- Advanced code completion with tree search
- Parallel comparison of hundreds of model responses
- Real-time content generation at scale
๐ฌ Core Specifications
- 144 GB HBM3E memory per chip
- Expansible to 100T parameter models
- Fully open-source software stack
- Outperforms NVIDIA 8xH100 and 8xB200 in LLaMA 70B throughput
Think of it as custom-built highways for AI instead of regular roads used by other computers
This is insane
My mouth wide open for all 5 minutes of this video.
This is an absolute game changer for 3D workflows.
BlenderGPT ๐คฏ๐ฅ๐ฑ Link below