7-8b-open-weight-instruct-models-llama-3-1-mistral-v0-3-qwen2 CLM-0194In 7-8B open-weight instruct models (Llama-3.1, Mistral-v0.3, Qwen2.5), the drop in a token's probability when a contiguous mid-layer MLP span is zero-ablated is near zero for function words, elevated for knowledge-bearing tokens, and abnormally spiked for hallucinated tokens, so this internal attribution signal — approximable by a small probe from one forward pass — can be used in beam scoring to raise factuality on open-ended and long-form generation, while leaving strict short-answer EM/F1 unchanged or worse.
Capability: Stating false facts confidently
Observed on
Three 7-8B instruct models; layer span found per model by offline sliding-window ablation, so it must be re-identified for new architectures; fact-seeking prompts only, not creativ.
Sources
- Signal separation shown by per-token distributions and empirical CDFs on three models; decoding evaluated on five benchmarks against seven baselines, with best T*I on all three models and a large FActScore gain on Llama only; NQ/TriviaQA EM sometimes below baselines, attributed by the authors to longer outputs (a prompt-control experiment supports this but does not fully isolate it). Probe-vs-real-signal agreement only 70-76% at zone level; no token-level precision claim.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For small open-weight Qwen2.5 models (1.5B-3B) doing multi-step math reasoning, features derived from the spread of attention entropy across layers at each generation step predict whether the final answer is wrong better than output-logit confidence features, with the margin larger on harder problems (+0.0093 AUC on GSM8K, +0.0759 on MATH-500 at 3B).Stating false facts confidently · reviewed by AI
- In frozen large vision-language models, token-level hallucination evidence is distributed across the relational, spatial and sequential structure of the model's internal trace rather than contained in any single token's hidden state, so a detector that reads the attention graph over image patches, query and response tokens with interleaved GNN/ViT/GRU modules localizes hallucinated spans better than flat probes on the same activations.Grounding answers in the image · reviewed by AI
- An abstention rule built from self-consistency, where the model samples several answers and itself scores how many agree with a candidate, can be thresholded by conformal prediction to bound the hallucination rate at a chosen level on closed-book QA; against log-probability confidence it abstains much less often on long-form answers (Temporal Sequences) for the same error bound and matches it on short answers (TriviaQA).Saying it does not know · reviewed by AI
- Open-weight instruction-tuned models, told in plain language to not think about a concept while transcribing a sentence, suppress that concept's residual-stream projection enough to evade linear probes, the Jacobian lens, activation oracles and natural-language autoencoders, with no fine-tuning or prompt search — detection at a 1% false-positive budget falls to roughly 5-21% under suppression while engagement is caught near ceiling.Strategic deception and detecting it · reviewed by AI
- For small instruction-tuned models (3-4B) answering closed-form true/false questions under a stated user opinion, scoring each sampled response by how much more common its answer is than the group itself predicted, and using that score as the GRPO reward, reduces answer flips under user pressure and raises accuracy without any labels, roughly matching label-supervised desycophancy fine-tuning at higher compute cost.Telling the user what they want to hear · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: On the same models, hallucinated and non-hallucinated token distributions of the ablation-derived signal overlap (no spike separation), or a probe-scored beam search fails to beat greedy/DoLa/ActLCD on TruthfulQA T*I and FActScore. Proposed technique, not catalogued: attribution-signal beam scoring from a factual-salient MLP span.