Prime Intellect Releases Verifiers v1: Composable Tasksets, Harnesses, and Runtimes for Agentic RL Training and Evaluations
Prime Intellect announced Verifiers v1 (verifiers 0.2.0), a framework that splits RL environments into tasksets, harnesses, and runtimes, with an interception server for trace recording.
- Verifiers v1 introduces a modular split of RL environments into tasksets, harnesses, and runtimes.
- An interception server records training traces automatically, easing data collection.
- The framework is compatible with prime‑rl, allowing immediate use for agent training.
Prime Intellect released Verifiers v1, version 0.2.0, introducing a new namespace verifiers.v1 that reorganizes reinforcement‑learning environments into three interchangeable components: tasksets (what to do), harnesses (how to execute), and runtimes (where it runs). This modular design allows any compatible harness to run any taskset, simplifying experimentation.
An interception server sits between the agent and the environment, proxying requests and automatically recording training‑ready traces. The system supports prime‑rl training out of the box, enabling researchers to generate reproducible datasets without custom logging code.
The release is positioned as a preview of a larger "v1" core, aiming to accelerate development of agentic RL systems by providing a plug‑and‑play infrastructure. While the toolkit is currently in early preview, it targets developers building custom RL pipelines and researchers needing standardized evaluation pipelines.
Provides a reusable, composable stack for building and testing RL agents.
Offers a hands‑on platform to study modular RL system design.
Simplifies creation of reproducible RL experiments for the broader AI community.
- taskset
- A definition of the specific RL task or goal the agent must achieve.
- harness
- The execution layer that determines how the taskset is run, such as simulation speed or API.
- runtime
- The environment or platform where the RL episode is executed.
- interception server
- A proxy that captures interactions between agent and environment to generate training traces.
Google DeepMind launches free 56-hour AI course in India | Indian startups get access to Gemini 3.5 Flash | Inshorts - Inshorts
AI ToolsPrismML Releases Bonsai 27B: 1-bit and Ternary Builds of Qwen3.6-27B That Run on Laptops and Phones
AI ToolsI Put a Hailo 8 in a Handheld and Stopped Paying for Inference
AI ToolsOpenCoreDev Releases Domain SDK 0.2.0: One TypeScript API to Add, Verify, and Remove Customer Domains Across Five Platforms
lobste.rs is now running on SQLite
AI In Chip Design: Lots Of Promise, Plenty Of Unanswered Questions - Semiconductor Engineering
AI In Chip Design: Lots Of Promise, Plenty Of Unanswered Questions Semiconductor Engineering
Nokia to Sell Nvidia-Powered AI Mobile Network Gear From 2027 - Bloomberg.com
Nokia announced it will sell mobile network gear powered by Nvidia AI chips starting in 2027 to enhance performance and efficiency.
HardwareOpenAI's first hardware product is a screenless AI speaker designed to feel alive
OpenAI is developing a portable, screenless smart speaker with a camera and moving parts, designed to feel alive. The launch is planned for 2027 but may be delayed due to a trade secrets lawsuit.
ASML raises 2026 forecast, expands capacity on AI chip demand - SRN News
ASML announced it has raised its 2026 revenue forecast and will increase production capacity to meet growing demand for AI chips.
Chip-Machine Supplier ASML Raises Guidance Again on Unrelenting AI Demand - WSJ
ASML, the Dutch lithography equipment maker, raised its 2024 revenue guidance again, citing relentless demand from AI-driven semiconductor production.
26 Meta employees sue, alleging AI-driven layoff picks hit workers on medical and parental leave - AP News
Twenty-six Meta employees have filed a lawsuit claiming the company's AI-driven layoff algorithm disproportionately targeted workers on medical or parental leave.