Gemma 4, small models fail basic grammar checks

// 109d agoNEWS

Gemma 4, small models fail basic grammar checks

Users report that recent open-weights models like Gemma 4 E4B and Qwen 3 Next fail at basic grammar checking by over-correcting valid text. The issue highlights a persistent fluency bias and prompt brittleness in smaller parameter architectures.

// ANALYSIS

Despite high reasoning benchmarks, small models remain surprisingly bad at surgical text editing because they are trained to rewrite rather than correct.

–Gemma 4 E4B "corrects" valid words to themselves, demonstrating an inability to execute a simple no-op when text is already correct
–Instruct-tuning conditions these models to always be "helpful," leading them to force unnecessary edits and add conversational filler
–Smaller architectures struggle to maintain strict negative constraints (like "do not change the meaning") across longer sentences
–The over-correction issue is widespread across new releases, affecting the recently launched GPT-OSS 20B and Qwen 3 Next 80B models

// TAGS

gemma-4llmopen-weightsreasoning

DISCOVERED

109d ago

2026-04-13

PUBLISHED

109d ago

2026-04-13

RELEVANCE

8/ 10

AUTHOR

Character-Blood3482

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE25m ago

Genspark Workspace 6.0 drops six major updates

Genspark Workspace 6.0 expands Genspark's ecosystem across six core updates designed to bridge ambient work context into executable workflows. Key releases include SecondBrain Note hardware voice recorder, GenTeam multi-agent collaboration, GenMail email workflows, Genspark Design, AI Slides, and AgentBase for custom databases.

NEWS29m ago

Google begins active development on Gemini 4

Google is reportedly actively developing Gemini 4, its next-generation foundation model designed to be its most advanced AI system to date. Key objectives for the new model include superior reasoning skills, improved coding assistance, and enhanced agentic capabilities for autonomous task execution, while Gemini 3.5 Pro continues testing behind the scenes.

RESEARCH1h ago

MANTA enables dynamic topology adaptation for multi-agent systems

MANTA (Multi-Agent Network Topology Adaptation) is a research framework that allows multi-agent LLM systems to dynamically reconfigure their communication topologies at inference time. By combining trace auditing with verbal playbooks during execution, it enables agent teams to optimize collaboration efficiency and achieve superior results on complex benchmarks such as PlanCraft.

Gemma 4, small models fail basic grammar checks