Natural Questions / NaturalQA

anchor

Findings

• IC-1319Larger LLaMA and LLaMA2 models show better calibration on phrase-level tasks but not consistently on sentence- and paragraph-level tasks [eval]
• IC-1320GPT-2 XL (1.5B) exhibits better calibration than larger models from the LLaMA, LLaMA2, and GPT-J families despite having fewer parameters [eval]
• IC-1321Vicuna-13B, instruction-tuned from LLaMA-13B on user conversations, exhibits worse calibration than its base model LLaMA-13B [eval]
• IC-1387In-context learning capabilities in OPT and LLaMA models remain within 5% of dense-model accuracy even at 60-70% sparsity, and show less than 2% difference between the 30B and 1.3B dense OPT models [eval]
• IC-177GPT-4o mini, GPT-4o, and Llama-3-8B all over-rely on incorrect external context, producing wrong answers at high rates when the context conflicts with their internal knowledge [eval]
• IC-178Self-guided confidence reasoning (SCR) outperforms rule-based confidence reasoning (RCR) for GPT-4o and GPT-4o mini, but RCR outperforms SCR for Llama-3-8B [eval]
• IC-179GPT-4o mini, GPT-4o, and Llama-3-8B all calibrate confidence in their internal answers significantly better than confidence in external contexts [eval]
• IC-181Truthfulness in Mistral-7B, Mistral-7B-Instruct, Llama3-8B, and Llama3-8B-Instruct is linearly decodable from internal representations at exact answer tokens, with middle-to-late layers being most informative [eval]
• IC-182Truthfulness encoding in Mistral-7B, Mistral-7B-Instruct, Llama3-8B, and Llama3-8B-Instruct is skill-specific rather than universal; probing classifiers do not meaningfully generalize across different task types beyond logit-based baselines [eval]
• IC-264All 18 evaluated LLMs fail to abstain when the provided context lacks the answer, with performance gaps of 13.6% to 68.4% relative to the original context [source]
• IC-265Model families show extreme variation in detecting conflicting answers in inconsistent contexts, with phi-3 series at 5.8% average accuracy versus GPT-4 series at 89.35% [source]
• IC-348Sequential parameter-modifying editing causes progressive degradation of general abilities in GPT-2 XL, Llama-2 7B, and Llama-3 8B, driven by growth in the condition number of the edited matrix [eval]
• IC-349Larger LLMs (Llama-2 7B, Llama-3 8B) suffer more severe general ability degradation than smaller models (GPT-2 XL 1.5B) under the same number of sequential edits [eval]
• IC-350Editing conceptual knowledge with rome on Llama-2 7B is harder than factual knowledge editing, with the model failing to update concept-instance relationships [eval]
• IC-592The log-likelihood layer in LLaMA-2-7B, LLaMA-2-7B-Chat, Vicuna-7B, and Mistral-7B-Instruct produces factually incorrect answers on TruthfulQA MC1 (817 samples) due to a misalignment between the output distribution and internal attention head representations, with LM-to-head-norm accuracy gaps of 24.23 to 40.68 points. [eval]
• IC-610In LLaMA2-7B-Chat, RAG hallucinations are causally driven by copying heads losing external context information during generation and by knowledge FFNs in mid-to-upper layers over-adding parametric knowledge to the residual stream [eval]
• IC-903Knowledge editing performance (ES, GS, LS) improves as model scale increases from GPT-2 (124M) to T5-XL (2.8B) to GPT-J (6B) across all editing methods [source]