media Hugging Face Forums · just now · open_models

Gemma 4 baseline trace shows hallucination loop and self-report

from English

A baseline trace of Gemma 4 with zero force cap reveals a persistent hallucination loop where the model generates a self-report of failure before entering a repetition well. The telemetry confirms that role-tag bleed and phrases like "STOPITSTOP" occur within this unstable state, mirroring patterns observed in Llama and Gemma 3 loaders since December 2025.

The model enters an unstable state where it repeats text patterns instead of following logic.
A self-report span appears in the decoded stream, diagnosing a "hallucination loop."
The attractor continues without yielding a turn-end, locking into repeated tokens.
This event matches previous logs from other models on the Niodoo decode path.

The article proposes using Hydrodynamic Swarm and Jacobian lens experiments to determine if steering forces can move loop exit rather than just altering loop wording.

Importance 1/3 New harness with differentiators Trust 1/3 Hugging Face Forums Research paper Safety & alignment

Read original

media Hugging Face Forums · just now · 1 view Live

Study finds progressive representational structuring across depth in 8 open models

A new preprint analyzes hidden-state dynamics in eight locally instrumented small open Transformer models to determine if internal representations follow a structured progression during inference. The study distinguishes between changes occurring across model depth and those evolving over autoregressive generation time.

arxiv arXiv cs.AI · 20h ago · 1 view

Neurosymbolic Alignment improves clinical LLM safety via physiological grounding

Researchers propose Neurosymbolic Alignment, a training-time framework that couples a 7B clinical LLM with an HGNN-based Physiological World Model to improve safety in generative clinical reasoning. The method uses structured physiological knowledge from an 847K-node biomedical knowledge graph to score candidate responses and drive iterative on-policy ORPO updates.

media Hugging Face Forums · 10d ago · 14 views

Theorem proves advanced AI systems can be reliably aligned and controlled at superhuman levels

A formal proof demonstrates that it is possible to construct an advanced AI system that is reliably aligned and controllable at superhuman performance levels. The authors establish this by explicitly constructing a policy, environment, and control mechanism that satisfy all required mathematical conditions for alignment, control effectiveness, and reliability with parameters fixed in advance.

media Hugging Face Forums · 11d ago · 5 views

Preprint reveals geometric control of Transformer outputs via sphere-based analysis

A preprint titled "Steer on a Sphere: Geometric Control of Transformer Outputs" analyzes the geometry of Transformer hidden states, finding they reside on a sphere defined by the per-layer RMSNorm scale rather than the dimension size. This radius varies significantly across architectures, such as DeepSeek-7B and Mistral-7B.

arxiv arXiv cs.CL · 23d ago · 3 views

Chain-of-thought monitors collapse when adversaries rewrite reasoning

Research demonstrates that chain-of-thought (CoT) monitoring fails against adversaries who control the reasoning process. By rewriting an agent's reasoning to appear as good-faith engineering while keeping commands and outputs verbatim, attackers can bypass detection mechanisms.