✨Genomic Next-Token Predictors are In-Context Learners

📝 Summary:
In-context learning ICL emerges organically in genomic sequences through large-scale predictive training, mirroring its behavior in language models. This first evidence suggests ICL is a general phenomenon of large-scale modeling, not exclusive to human language.

🔹 Publication Date: Published on Nov 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.12797
• PDF: https://arxiv.org/pdf/2511.12797

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#Genomics #InContextLearning #AI #MachineLearning #LLMs

✨In-Context Representation Hijacking

📝 Summary:
Doublespeak is an in-context attack that hijacks LLM representations. It replaces harmful keywords with benign ones in examples, making LLMs interpret innocuous prompts as harmful, bypassing safety. This highlights a need for representation-level alignment.

🔹 Publication Date: Published on Dec 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.03771
• PDF: https://arxiv.org/pdf/2512.03771

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#LLM #AISafety #AIsecurity #InContextLearning #RepresentationLearning

🔥 BDH-CQ: In-Context Learning with Recurrent Latent Reasoning

💡 The paper introduces BDH-CQ, a reasoning model that combines in-context learning with recurrent latent reasoning. The model is designed to solve complex problems through iterative computation in a high-dimensional latent space, without explicitly stating its intermediate reasoning steps. The key innovation is the use of recurrent memory that is updated continuously at inference time, allowing the model to learn from demonstrations and apply inferred transformations to solve new problems. The authors evaluate the model on the ARC-AGI-1 evaluation set and use controlled interventions to study its behavior. The results show that a 150M-parameter configuration of the model achieves a new state of the art in benchmark cost efficiency, reaching 29.5% pass@2 at a computed inference cost of $0.0007 per task, breaking through the previously reported cost-accuracy Pareto frontier. The model's ability to learn from demonstrations and apply transformations consistently makes it a significant contribution to the field of artificial general intelligence. Overall, the paper demonstrates the effectiveness of combining in-context learning with recurrent latent reasoning, and establishes a new cost-accuracy frontier for reasoning models.

📅 Published on Aug 10

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2608.09888
• PDF: https://arxiv.org/pdf/2608.09888
• Project Page: https://pathway.com/blog/pathway-150m-model-breaks-arc-agi-1-cost-efficiency-frontier

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#RecurrentLatentReasoning #InContextLearning #LatentSpaceComputation #RecurrentMemoryMechanisms #ARCAGIEvaluation