"there's nothing interesting on arxiv these days!" - the words of...

"there's nothing interesting on arxiv these days!"
- the words of an uncurious mind

i have personally been blown away by the volume of interesting papers posted over the last few months, and eagerly following daily digests

here are some papers i enjoyed the most:

- Pre-training under infinite compute (September 2025, arxiv.org/abs/2509.14786)
- Fresh in memory: Training-order recency is linearly encoded in language model activations (September 2025, arxiv.org/abs/2509.14223)
- Subliminal Learning: Language models transmit behavioral traits via hidden signals in data (July 2025, arxiv.org/abs/2507.14805)
- Memory Limitations of Prompt Tuning in Transformers (September 2025, arxiv.org/abs/2509.00421)
- Behavioral Fingerprinting of Large Language Models (September 2025, arxiv.org/abs/2509.04504)
- Language Self-Play For Data-Free Training (September 2025, arxiv.org/abs/2509.07414)
- The Illusion of Diminishing Returns: Measuring Long Horizon Execution in LLMs (September 2025, arxiv.org/abs/2509.09677)
- Do Natural Language Descriptions of Model Activations Convey Privileged Information? (September 2025, arxiv.org/abs/2509.13316)
- Beyond the Leaderboard: Understanding Performance Disparities in Large Language Models via Model Diffing (September 2025, arxiv.org/abs/2509.18792)
- Stochastic activations (September 2025, arxiv.org/abs/2509.22358)
- PonderLM-2: Pretraining LLM with Latent Thoughts in Continuous Space (September 2025, arxiv.org/abs/2509.23184)
- Words That Make Language Models Perceive (October 2025, arxiv.org/abs/2510.02425)
- Language Models Do Not Embed Numbers Continuously (October 2025, arxiv.org/abs/2510.08009)
- Learning Facts at Scale with Active Reading (August 2025, arxiv.org/abs/2508.09494)
- OverFill: Two-Stage Models for Efficient Language Model Decoding (August 2025, arxiv.org/abs/2508.08446)
- Retrieval Capabilities of Large Language Models Scale with Pretraining FLOPs (August 2025, arxiv.org/abs/2508.17400)
- Reasoning-Intensive Regression (August 2025, arxiv.org/abs/2508.21762)
- Watch the Weights: Unsupervised monitoring and control of fine-tuned LLMs (August 2025, arxiv.org/abs/2508.00161)
- On the Theoretical Limitations of Embedding-Based Retrieval (August 2025, arxiv.org/abs/2508.21038)