Hedra has unveiled Model Factory, a post-training and inference platform built specifically for physical AI. The system uses recursive self-improvement to adapt existing models to new tasks targeting robotics, manufacturing, biology and defense teams. Customers supply base models and proprietary data while Hedra researchers orchestrate training runs using vision-language models and reinforcement learning. Results can be deployed on premises or distilled for edge devices.
The company reports impressive speed gains on AMD hardware. Hedra claims a 6x inference speedup on identical AMD GPUs, and Krea 3 image edits dropped from 22 seconds to 3.6 seconds when running on four AMD Instinct MI355X GPUs. In one simulation, Hedra researchers restored a robot's task success from near zero to near 100% after a gripper swap, demonstrating the platform's ability to retrain models for changed physical conditions.
Lab Thread, a UK-based life science software company, launched a Model Context Protocol server that lets scientists connect their own AI assistants directly to laboratory data. The tool is designed to work with a wide range of AI assistants, helping researchers accelerate their work and improve findings. Separately, Vizgen created Velocity Lab, a new business unit that will generate large-scale AI training data for spatial data production, serving both external customers and Vizgen's own needs.
On the policy and social front, experts advise workers returning after AI-related layoffs to negotiate higher pay, request one-year renewable contracts with buyout clauses, and emphasize human skills like empathy and creativity that AI cannot easily replicate. Meanwhile, mathematicians and educators are pushing back on assumptions about AI's role in complex fields. Steven Strogatz argues math is a way of thinking that AI cannot replace, while Rutgers Prof. Britt Paris told a Northwestern audience that AI is brittle and not inevitable, pointing to community broadband efforts in Philadelphia and Chattanooga as proof that communities can take control of technology at scale.
Key Takeaways
- Hedra's Model Factory platform uses recursive self-improvement to retrain physical AI models for robotics, manufacturing, biology and defense
- Hedra reports a 6x inference speedup on identical AMD GPUs and cut Krea 3 image edit time from 22 seconds to 3.6 seconds on four AMD Instinct MI355X GPUs
- Hedra restored a robot's task success from near zero to near 100% in simulation after a gripper swap using the Model Factory
- Lab Thread launched a Model Context Protocol server enabling scientists to connect AI assistants directly to laboratory data
- Vizgen launched Velocity Lab, a new business unit generating large-scale AI training data for spatial data production
- Experts recommend workers returning after AI layoffs negotiate higher pay and secure contracts with buyout clauses
- Doctors warn patients often miss errors and hallucinations in AI-generated medical findings and should share them with physicians for review
- Mathematician Steven Strogatz argues AI cannot replace human intuition and creativity in mathematics
- Rutgers Prof. Britt Paris says AI is brittle and not inevitable, citing community broadband efforts in Philadelphia and Chattanooga
- An Austrian-school analysis frames the AI boom as a monetary policy-driven bubble, with slowdowns in money-supply growth expected to starve misallocated projects
Hedra claims 6x speedup for real-world models
Hedra launched Model Factory, an autonomous post-training and inference platform. The company reported a 6x inference speedup on identical AMD GPUs. Krea 3 image edits dropped from 22 seconds to 3.6 seconds on four AMD Instinct MI355X GPUs. The platform targets robotics, manufacturing, biology and defense teams. Hedra researchers also restored a robot's task success from near zero to near 100% in simulation after a gripper swap.
Hedra's new Model Factory cranks up production for physical AI model training
Hedra announced the Model Factory, a post-training and inference platform designed for physical AI. It uses recursive self-improvement to adapt existing models to new tasks for robots and machines. The platform serves robotics, manufacturing, biology and defense customers. Customers provide base models and proprietary data, while Hedra researchers orchestrate training runs using vision-language models and reinforcement learning. Models can be deployed on premises or distilled for edge devices.
3 AI Architecture Tools Rethinking Workflows
Three AI tools are changing how architects work. xFigura brings image, video, 3D and other models into one shared canvas with three layers of intelligence. Motif aims to replace Revit while understanding existing Revit and Rhino models in a web-based environment. QikBIM focuses on automating drawing-to-model conversion. These tools target different points in architecture's fragmented digital workflow.
Is AI A Financial Bubble?
An Austrian-school analysis argues the AI boom is a classic bubble caused by expansionary monetary policy. Low interest rates created by credit expansion made otherwise unprofitable AI projects appear viable. The real affordability depends on the savings fund, not technology strength. A slowdown in money-supply growth since early 2021 is expected to starve misallocated activities of resources. Further rate cuts would deepen the eventual bust.
Lab Thread launches new AI tools and strengthens platform with universal AI assistant compatibility
Lab Thread Ltd, a UK-based life science software company, launched a Model Context Protocol server. The new tool lets scientists connect their own AI assistants directly to laboratory data. It is designed to integrate with a wide range of AI assistants. The company says this will help scientists leverage AI to accelerate research and improve findings.
Vizgen Creates Velocity Lab for AI Spatial Data
Vizgen launched Velocity Lab, a new business unit that will absorb the company's existing service lab. It will generate large-scale AI training data for external customers and the company itself. The goal is to speed up AI model development for spatial data production. This should help Vizgen deliver more accurate and efficient results.
Advice for Workers Asked Back After AI Job Cuts
Experts say workers returning after AI-related layoffs should ask for higher pay and clarity on how AI affects future staffing. They recommend getting a one-year renewable contract with a buyout clause. Workers should also highlight human skills like empathy, creativity and team building that AI cannot easily replace. Asking for training and a clear advancement plan can also help.
Relying on AI for Medical Diagnoses Can Be Risky
Patients are increasingly bringing AI-generated medical findings to doctor appointments and trusting them more than internet searches. Doctors say patients often do not notice errors or hallucinations in AI output. Experts encourage patients to share AI findings with physicians so they can help filter inaccuracies. Using AI to record medical visits also carries risks because the notes may contain errors that need physician review.
Mathematician Discusses AI and Big Math Book
Mathematician Steven Strogatz discussed his book Big Math in an interview about AI and mathematics. He argues that math is a way of thinking, not just a problem-solving tool. While AI can automate tasks and aid discovery, Strogatz says it cannot replace human intuition and creativity. The book explores how math helps us understand complex systems and the world around us.
Rutgers Professor Challenges AI Inevitability at NU
Rutgers Prof. Britt Paris told a Northwestern audience that AI is not inevitable and is rather brittle. She argued that digital systems reflect choices about ownership and labor that communities can challenge. Paris pointed to community broadband efforts in Philadelphia and Chattanooga as examples. She said technology at scale is possible if communities control it, and history shows the status quo will not last forever.
Sources
- Hedra claims 6x speedup for real-world models
- Hedra's new Model Factory cranks up the production volume for physical AI model training
- 3 AI Architecture Tools Rethinking Workflows
- Is AI A Financial Bubble?
- Lab Thread launches new AI tools and strengthens platform with universal AI assistant compatibility
- Vizgen Launches New Business Unit for AI Model Spatial Data Production
- What to consider if a company that made AI-related job cuts wants you back
- Why Relying on AI for Medical Diagnoses Can Be Risky
- What happens to mathematics when AI becames an oracle?
- Rutgers professor challenges AI’s ‘inevitability’ during NU lecture
Comments
Please log in to post a comment.