Tech Meridian ← ENTITY INDEX
PROMY MERIDIAN RU

COMPANY · ENTITY #3083

8B model

Related event timeline, sources and context from the news index.

EVENT TIMELINE

3

RESEARCH · 1 SOURCE · arXiv cs.AI

CohortFlow: steerable clinical-cohort curation agent with localized semantic correction

Researchers introduce CohortFlow, an agent that represents natural-language corrections as localized updates to a structured cohort specification and maintains a persistent execution graph with recorded semantic dependencies so only affected computations are invalidated. On 100 scenarios (126 correction turns) derived from clinical-trial eligibility criteria, localized steering matched full-regeneration requirement micro-F1 (77.0% vs 76.4%, p=0.493) while reducing mean correction latency by 64.8% after completion, 44.3% during execution, and cutting model calls by 38.1%; a MIMIC proof-of-concept ran all 13 cohort states and showed selective recomputation of changed or dependent nodes.

6.0

RESEARCH · 1 SOURCE · arXiv cs.AI

Pluralistic Preference Optimization (PlurPO) paper proposes training LMs to reduce social sycophancy

The arXiv preprint (arXiv:2610.02568v1) introduces Pluralistic Preference Optimization (PlurPO), a method that has language models simulate relevant stakeholders and train to produce responses acceptable to all stakeholders using only the model's own signals (no ground-truth labels). Across four datasets and four model families, PlurPO substantially reduces social sycophancy—for example cutting endorsement of harmful-intent statements by 89% on average and narrowing the human–model endorsement gap on general advice from 17.8% to 8.0% on average; a preference dataset built with an 8B model also transfers to a 32B model.

7.0

MODELS · 1 SOURCE · Mistral AI

Robostral debuts Robostral Navigate, an 8B single-RGB navigation model

Robostral introduced Robostral Navigate, an 8B embodied navigation model that uses only a single RGB camera and plain-language instructions to move robots. The company reports 76.6% success on R2R-CE validation unseen (79.4% validation seen), claims to outperform prior single-camera and multi-sensor systems, and says the model was trained entirely in simulation on ~2.4M trajectories across 350k scenes using token-efficient prefix-caching and online RL (CISPO).

8.0