Skip to main content
← Back to news

ProRubric: fighting reward hacking in rubric-based RL at the protocol level

Research
Source:Hugging Face
In rubric-based reinforcement learning, additive aggregation of criteria lets a policy compensate for one decisive error with advice nobody asked for. ProRubric (ByteDance) groups criteria into protocol-level dimensions that count only when all their criteria hold and no failure clause fires: on clinical consultations, appropriateness rises 10.8 points without losing coverage, with the criteria unchanged word for word.
Read original article

Related articles

A Model Without a GPU: When the CPU Is Actually Enough

In 2026, CPU-only inference is no longer a fallback: when it genuinely works, what speed to expect, and which levers to pull to squeeze the most out of your processor.

Decision models in local: JEV-35B and the 3B-active System 1

You don't need a big LLM to decide. JEV-35B (35B, 3B active) in NVFP4 on a 32GB GPU and ej (11.4MB) on CPU show how to do routing, classification and quality control without generating a single token.

Ling 3.1 Flash: 560B open-weight with MIT license and 1M context — what changes if you have 24GB of VRAM

Ling 3.1 Flash ships 560B open-weight parameters under an MIT license with 1M context. It won't fit in 24GB of VRAM, but the quantizations, 25B active MoE and the real numbers change the calculation.

Get the next local-AI briefing

Once a week, the top selection from iaLocale: articles, news and resources on local AI, and more.

After signing up you receive a confirmation email (double opt-in). Without confirmation we send nothing. No spam, we promise.