Tootfinder

IFDECORATOR: Wrapping Instruction Following Reinforcement Learning with Verifiable Rewards
Xu Guo, Tianyi Liang, Tong Jian, Xiaogui Yang, Ling-I Wu, Chenhui Li, Zhihui Lu, Qipeng Guo, Kai Chen
https://arxiv.org/abs/2508.04632

IFDECORATOR: Wrapping Instruction Following Reinforcement Learning with Verifiable Rewards
Reinforcement Learning with Verifiable Rewards (RLVR) improves instruction following capabilities of large language models (LLMs), but suffers from training inefficiency due to inadequate difficulty assessment. Moreover, RLVR is prone to over-optimization, where LLMs exploit verification shortcuts without aligning to the actual intent of user instructions. We introduce Instruction Following Decorator (IFDecorator}, a framework that wraps RLVR training into a robust and sample-efficient pipeline…

#programminghumor #memes

Story links:
https://arstechnica.com/security/2025/09/former-whatsapp-security-boss-sues-meta-for-systemic-cybersecurity-failures/
https://arstechnica.com/google/2025/09/in-court-filing-google-concedes-the-open-web-is-in-rapid-decline/

Anyway, I was looking really fresh and feeling calm, just flowing with good energy earlier today.
#Nature #Relaxing #Zen

I am feeling oddly competitive now that my twelve year old has caught up with my progress in Silksong.
#Silksong

Filing: Nasdaq asks the US SEC to let investors trade tokenized securities on its exchange, potentially rolling out the first token-settled trades in Q3 2026 (Anirban Sen/Reuters)
https://www.reuters.com/business/finance/nasdaq-makes…

NFL Power Rankings, Week 6: Lions leapfrog Bills and Eagles to claim No. 1 spot; Chargers keep falling https://www.nfl.com/news/nfl-power-rankings-week-6-2025-nfl-season

Results on long twins in random words and permutations
Elliott Liu, Linus Tang, Jessica Wan
https://arxiv.org/abs/2510.04335 https://arxiv.org/pdf/2510.043…

Results on long twins in random words and permutations
We study long $r$-twins in random words and permutations. Motivated by questions posed in works of Dudek-Grytczuk-Ruciński, we obtain the following. For a uniform word in $[k]^n$ we prove sharp one-sided tail bounds showing that the maximum $r$-power length (the longest contiguous block that can be partitioned into $r$ identical subblocks) is concentrated around $\frac{\log n}{(r-1)\log k}$. For random permutations, we prove that for fixed $k$ and $r\to\infty$, a uniform permutation of $[rk]$ …

Backdoor-Powered Prompt Injection Attacks Nullify Defense Methods
Yulin Chen, Haoran Li, Yuan Sui, Yangqiu Song, Bryan Hooi
https://arxiv.org/abs/2510.03705 https://

Backdoor-Powered Prompt Injection Attacks Nullify Defense Methods
With the development of technology, large language models (LLMs) have dominated the downstream natural language processing (NLP) tasks. However, because of the LLMs' instruction-following abilities and inability to distinguish the instructions in the data content, such as web pages from search engines, the LLMs are vulnerable to prompt injection attacks. These attacks trick the LLMs into deviating from the original input instruction and executing the attackers' target instruction. Recently, var…

Knowledge Collapse in LLMs: When Fluency Survives but Facts Fail under Recursive Synthetic Training
Figarri Keisha, Zekun Wu, Ze Wang, Adriano Koshiyama, Philip Treleaven
https://arxiv.org/abs/2509.04796

Knowledge Collapse in LLMs: When Fluency Survives but Facts Fail under Recursive Synthetic Training
Large language models increasingly rely on synthetic data due to human-written content scarcity, yet recursive training on model-generated outputs leads to model collapse, a degenerative process threatening factual reliability. We define knowledge collapse as a distinct three-stage phenomenon where factual accuracy deteriorates while surface fluency persists, creating "confidently wrong" outputs that pose critical risks in accuracy-dependent domains. Through controlled experiments with recursiv…