Gabriele Merlin
@gab709_1
Followers
36
Following
4
Media
1
Statuses
10
Joined January 2024
RT @MathisPink: 1/n🤖🧠 New paper alert!📢 In "Assessing Episodic Memory in LLMs with Sequence Order Recall Tasks" ( w….
arxiv.org
Current LLM benchmarks focus on evaluating models' memory of facts and semantic relations, primarily assessing semantic aspects of long-term memory. However, in humans, long-term memory also...
0
17
0
6/6.🚀Additionally, fine-tuning the model on specific stimulus texts can improve alignment with brain recordings corresponding to held-out text!.
1
0
1
5/6.🎉 We show that the alignment between language models and brain recordings in regions like the inferior frontal gyrus (IFG) and angular gyrus (AG) is influenced by more than just next-word prediction and word-level information and it may be due to multi-word information.
1
0
2
4/6.✨ Our approach applies perturbations to language models, then contrasts the predicted brain activity across different model variations to disentangle the factors that lead to a better alignment.
1
0
1
3/6.🔍 In our work, we disentangle the roles of next-word prediction and word-level information from other factors—like multi-word information—when analyzing the alignment between GPT-2 models and brain recordings.
1
0
1
2/6.🤔 Is next-word prediction necessary for this alignment, or is it just sufficient? Could other types of information be similarly important for brain-LM alignment?.
1
0
0
1/6.📖 Language models pretrained to predict the next word across billions of text documents can significantly predict brain activity in people comprehending language. This alignment is thought to be driven by next-word prediction as a key mechanism (Goldstein et al., 2022).
1
0
0
🚨 Thrilled to announce that our paper "Language models and brains align due to more than next-word prediction and word-level information" has been accepted at #EMNLP2024! 🧠🤖.Work done in collaboration with @mtoneva1. Curious about why brain and GPT-2 models align? .Read on👇
3
7
43