{"slug": "the-capacity-of-thought-benchmarking-llama-3-2-in-semantic-fmri-neural-language", "title": "The Capacity of Thought: Benchmarking Llama 3.2 in Semantic fMRI Neural Language Decoding and Improving the Huth Encoding-Model Baseline", "summary": "Researchers benchmarked Llama 3.2 for semantic fMRI neural language decoding and improved the Huth encoding-model baseline. They achieved an 11% relative METEOR gain by enhancing the ridge regression pipeline, but found that a high-capacity language model (fMRIFlamingo with Llama-3.2-1B) produced near-identical scores with zeroed fMRI inputs, indicating decoding success was driven by the language prior rather than neural input. The study highlights the need for rigorous blind-control evaluation in fMRI decoding.", "body_md": "arXiv:2607.12079v1 Announce Type: new\nAbstract: Decoding continuous language from fMRI signals remains a core challenge in non-invasive brain-computer interface research. We present two complementary investigations. First, we improve the Huth et al. ridge regression encoding pipeline through expanded voxel selection (10K->15K), substitution of GPT-2 medium for GPT-1 as the beam-search proposal model, and GPU-accelerated bootstrap training, achieving mean METEOR = 0.149 and BLEU-1 = 0.200 across three held-out narratives for subject UTS03 -- an 11% relative METEOR gain over our replication baseline. Second, we introduce fMRIFlamingo, which maps BOLD activity to a frozen Llama-3.2-1B with trainable gated cross-attention layers via a learned brain tokenizer and a Perceiver Resampler. Despite achieving 42.86% Top-1 accuracy on a 1-in-100 ranking task, well above chance, a blind control ablation with zeroed fMRI inputs yields near-identical scores, revealing that apparent decoding success is driven primarily by the frozen language prior rather than by neural input. These results demonstrate that high-capacity language models do not inherently improve fMRI decoding and can actively obscure failures without rigorous blind-control evaluation.", "url": "https://wpnews.pro/news/the-capacity-of-thought-benchmarking-llama-3-2-in-semantic-fmri-neural-language", "canonical_source": "https://arxiv.org/abs/2607.12079", "published_at": "2026-07-15 04:00:00+00:00", "updated_at": "2026-07-15 04:32:36.959826+00:00", "lang": "en", "topics": ["large-language-models", "artificial-intelligence", "machine-learning", "neural-networks", "ai-research"], "entities": ["Llama 3.2", "Huth", "GPT-2", "GPT-1", "fMRIFlamingo", "UTS03", "METEOR", "BLEU-1"], "alternates": {"html": "https://wpnews.pro/news/the-capacity-of-thought-benchmarking-llama-3-2-in-semantic-fmri-neural-language", "markdown": "https://wpnews.pro/news/the-capacity-of-thought-benchmarking-llama-3-2-in-semantic-fmri-neural-language.md", "text": "https://wpnews.pro/news/the-capacity-of-thought-benchmarking-llama-3-2-in-semantic-fmri-neural-language.txt", "jsonld": "https://wpnews.pro/news/the-capacity-of-thought-benchmarking-llama-3-2-in-semantic-fmri-neural-language.jsonld"}}