Hey there, friend using a screen reader. I've carefully labeled every button and image on this site. Hope you have a smooth experience. If anything is inconvenient, my email is in the footer — just let me know and I'll fix it. — Store Owner
English, papers, AI, events, open source, contests, internship and graduate roles, and admissions. Choose what matters today.
Source titles, descriptions and official notices are shown in their original language. Some content is in Chinese.
This edition updated · Beijing time
First publication starts at 08:40 Beijing time, with more content added later. · Read the previous edition
Beijing time · Last 30 days
01
Daily English
Edition 2026-10-30
食堂餐卡充值的小意外
学会在食堂充值时说明问题并礼貌请求帮助。
Original English practice generated automatically each day
Expressions
top upMeaning & example
充值
I need to top up my meal card before lunch.
午饭前我需要给餐卡充值。
run out ofMeaning & example
用完，耗尽
I ran out of money on my card, so I couldn't buy dinner.
我卡里的钱用完了，所以没法买晚饭。
sort outMeaning & example
解决，处理
The staff helped me sort out the problem with my card.
工作人员帮我解决了卡的问题。
A short read
On Friday morning, Li Ming went to the campus dining hall for breakfast. He had run out of money on his student meal card the night before, so he needed to top up the card first. He walked to the service window and told the staff member, “I’d like to top up my card with 100 yuan, please.” The staff member took his card and checked the machine. After a moment, she said, “Sorry, the card reader is not working well today. It may take a few minutes.” Li Ming was a little worried because his first class would start in twenty minutes. He asked politely, “Is there another way to top up?” The staff member showed him a self-service machine nearby and helped him sort out the problem. Li Ming thanked her and quickly bought a sandwich. He made it to class just in time.
Choose your answers, then submit to see the explanations.
02
Daily Papers
Edition 2026-10-30
每日论文精选
自动整理 3 篇近七天的 arXiv 预印本，保留原始摘要与来源日期。
01cs.LG · cs.CL · cs.CR· v1
Heads of Security: Learned Attention Gates Improve Prompt Injection Robustness
Preprint source · Verify publication and peer-review status in the original
Authors: Igor Shilov, Xiaoxue Yang, Alexander Panfilov, Evgenii Kortukov, Matthieu Meeus, Peter Romov, Wojciech Samek, Jonas Geiping, Yves-Alexandre de Montjoye
We introduce Heads of Security (HoS), a prompt injection defense that enforces instruction-data separation via a small set of learned per-head attention gates. HoS learns how each head should redistribute attention between trusted and untrusted tokens, enabling models to reason over untrusted content without following injected instructions. We evaluate HoS on Llama-3.3-70B-Instruct and GPT-OSS-20B. We compare it against SecAlign++, a state-of-the-art finetuning-based defense, and ASIDE, a defense that, like ours, builds instruction-data separation into the model architecture. We show that HoS offers greater robustness to adaptive attacks, better performance on standard utility benchmarks, and matches top performance on standard agentic prompt injection benchmarks. More specifically, the black-box RL-Hammer attack reaches 97% ASR against GPT-OSS-20B and 85% against SecAlign, but only 10% against HoS. Claudini, a state-of-the-art white-box attack, strengthened against HoS through autoresearch, requires 3 times more FLOPs to reach the same ASR on HoS as it does on GPT-OSS-20B or SecAlign. HoS preserves base-model utility across standard multiple-choice benchmarks and prompting setups, whereas SecAlign and ASIDE degrade in few-shot settings. On agentic benchmarks with static attacks, HoS and SecAlign perform similarly and substantially improve the security-utility trade-off over the base model. ASIDE struggles with tool use, achieving only 10% utility. These results show that prompt injection robustness can be achieved and improved through targeted attention interventions.
Show original abstract
We introduce Heads of Security (HoS), a prompt injection defense that enforces instruction-data separation via a small set of learned per-head attention gates. HoS learns how each head should redistribute attention between trusted and untrusted tokens, enabling models to reason over untrusted content without following injected instructions. We evaluate HoS on Llama-3.3-70B-Instruct and GPT-OSS-20B. We compare it against SecAlign++, a state-of-the-art finetuning-based defense, and ASIDE, a defense that, like ours, builds instruction-data separation into the model architecture. We show that HoS offers greater robustness to adaptive attacks, better performance on standard utility benchmarks, and matches top performance on standard agentic prompt injection benchmarks. More specifically, the black-box RL-Hammer attack reaches 97% ASR against GPT-OSS-20B and 85% against SecAlign, but only 10% against HoS. Claudini, a state-of-the-art white-box attack, strengthened against HoS through autoresearch, requires 3 times more FLOPs to reach the same ASR on HoS as it does on GPT-OSS-20B or SecAlign. HoS preserves base-model utility across standard multiple-choice benchmarks and prompting setups, whereas SecAlign and ASIDE degrade in few-shot settings. On agentic benchmarks with static attacks, HoS and SecAlign perform similarly and substantially improve the security-utility trade-off over the base model. ASIDE struggles with tool use, achieving only 10% utility. These results show that prompt injection robustness can be achieved and improved through targeted attention interventions.
Source: arXiv · Official source · Source date: Oct 28, 2026
Attention as Lattice Gauge Theory: Relations Interact Only Through Wilson Loops
Preprint source · Verify publication and peer-review status in the original
Authors: Aakash Sarkar, Ananya Mukherjee, Motoshige Sato
Relations compose through the entity they share - Alice is Bob's mother and Bob is Carol's father, so Alice is Carol's grandmother - and the order matters, since a father's mother is not a mother's father; verifying a composed relation means closing the loop. We show that if each entity's coordinate frame may be rotated independently (per-entity frame freedom), three things follow. In the pure-gauge sector, relations that share no cycle cannot interact at all. The only gauge-invariant score of a triangle is a class function of its holonomy $H = T_{ij} T_{jk} T_{ki}$ - a Wilson loop. And the only equivariant update of an edge is $U = Ψ(H)\, T_{ij}$ with $Ψ$ conjugation-equivariant. Edge attention by Wilson loops is therefore not one design among many but the unique form the symmetry permits. Three communities hold pieces of this form: the commuting-operator embeddings of knowledge graphs are its abelian (commuting) sector; AlphaFold's triangle update and the Edge Transformer have its shape - an edge updated through the third vertex - without its constraints; and lattice QCD builds it equivariantly on lattices. We instantiate the equivariant sub-family on general relational graphs as holonomy attention. With no training and composition alone, it infers unobserved $S_4$ relations on sparse graphs at 0.88 with a fifth of the edges corrupted, at or slightly below specialist synchronization algorithms, where every commutative architecture is capped at 0.083; its holonomy score flags corrupted triangles at AUC 0.99 where commuting scores are capped at 0.76 by a second selection rule, and it separates clusters without labels.
Show original abstract
Relations compose through the entity they share - Alice is Bob's mother and Bob is Carol's father, so Alice is Carol's grandmother - and the order matters, since a father's mother is not a mother's father; verifying a composed relation means closing the loop. We show that if each entity's coordinate frame may be rotated independently (per-entity frame freedom), three things follow. In the pure-gauge sector, relations that share no cycle cannot interact at all. The only gauge-invariant score of a triangle is a class function of its holonomy $H = T_{ij} T_{jk} T_{ki}$ - a Wilson loop. And the only equivariant update of an edge is $U = Ψ(H)\, T_{ij}$ with $Ψ$ conjugation-equivariant. Edge attention by Wilson loops is therefore not one design among many but the unique form the symmetry permits. Three communities hold pieces of this form: the commuting-operator embeddings of knowledge graphs are its abelian (commuting) sector; AlphaFold's triangle update and the Edge Transformer have its shape - an edge updated through the third vertex - without its constraints; and lattice QCD builds it equivariantly on lattices. We instantiate the equivariant sub-family on general relational graphs as holonomy attention. With no training and composition alone, it infers unobserved $S_4$ relations on sparse graphs at 0.88 with a fifth of the edges corrupted, at or slightly below specialist synchronization algorithms, where every commutative architecture is capped at 0.083; its holonomy score flags corrupted triangles at AUC 0.99 where commuting scores are capped at 0.76 by a second selection rule, and it separates clusters without labels.
Source: arXiv · Official source · Source date: Oct 28, 2026
KPPax: preprocess differentiable kinetic ODE for machine learning in atmospheric chemistry
Preprint source · Verify publication and peer-review status in the original
Authors: Wenqing Peng, Michael Boy, Zhi-Song Liu
Atmospheric chemical mechanisms are commonly distributed in the language of the Kinetic PreProcessor (KPP), which generates efficient Fortran or C code but does not fit into the differentiable programming frameworks in which machine-learning models are developed. We present KPPax, a kinetic preprocessor that generates differentiable, just-in-time compiled, and vectorisable chemistry code in JAX from mechanisms written in the KPP language. KPPax evaluates the rate law with sparse operations in log space, which keeps it differentiable, and integrates the stiff system with the implicit solvers and the forward- and reverse-mode adjoints of the \texttt{diffrax} library, so that exact sensitivities are obtained without hand-derived adjoint code. We demonstrate two applications with the POLLU benchmark mechanism. In the first, the Fisher information matrix computed from an automatically differentiated Jacobian predicts, before any fit, which of the 25 rate coefficients a flow-tube-like design can identify, and noisy synthetic estimation confirms the predicted uncertainties of the nine identifiable rate coefficients to within about 30\% while revealing clusters of fast reactions of which only ratios are constrained. In the second, a neural-network correction of an emission inventory is trained offline on per-scenario 4D-Var retrievals and then fine-tuned online through the chemical model, which reduces the error of the biogenic correction by a factor of 2.4--5 and recovers information on low-temperature emissions that the individual retrievals lose. KPPax makes KPP mechanisms directly usable for parameter estimation and hybrid physics--machine-learning modelling.
Show original abstract
Atmospheric chemical mechanisms are commonly distributed in the language of the Kinetic PreProcessor (KPP), which generates efficient Fortran or C code but does not fit into the differentiable programming frameworks in which machine-learning models are developed. We present KPPax, a kinetic preprocessor that generates differentiable, just-in-time compiled, and vectorisable chemistry code in JAX from mechanisms written in the KPP language. KPPax evaluates the rate law with sparse operations in log space, which keeps it differentiable, and integrates the stiff system with the implicit solvers and the forward- and reverse-mode adjoints of the \texttt{diffrax} library, so that exact sensitivities are obtained without hand-derived adjoint code. We demonstrate two applications with the POLLU benchmark mechanism. In the first, the Fisher information matrix computed from an automatically differentiated Jacobian predicts, before any fit, which of the 25 rate coefficients a flow-tube-like design can identify, and noisy synthetic estimation confirms the predicted uncertainties of the nine identifiable rate coefficients to within about 30\% while revealing clusters of fast reactions of which only ratios are constrained. In the second, a neural-network correction of an emission inventory is trained offline on per-scenario 4D-Var retrievals and then fine-tuned online through the chemical model, which reduces the error of the biogenic correction by a factor of 2.4--5 and recovers information on low-temperature emissions that the individual retrievals lose. KPPax makes KPP mechanisms directly usable for parameter estimation and hybrid physics--machine-learning modelling.
Source: arXiv · Official source · Source date: Oct 28, 2026
Current source coverage: Selected events from connected official announcements, including OpenAI, Qwen, Microsoft Research, Google DeepMind and GitHub (including the Universe topic RSS), with an AI and developer focus.
No newly confirmed events in this edition
The checked sources have no new, unselected content. You can read an earlier edition.
Current source coverage: Currently covers the official Codeforces programming schedule and confirmed contests within the next 30 days; it is not a directory of every competition.
1 场仍未开始的官方比赛，所列时间为开赛时间。
01
Kotlin Heroes: Practice 15
Contest starts: Nov 20, 2026, 02:30 PM (UTC)
Codeforces 官方未来赛程。所列为开赛时间；报名窗口、组别与参赛条件以官方页面为准。
Project / Employer:
Codeforces
Location and work mode:
线上 · 全球 / Online · Worldwide
Requirements:
注册及组别资格请以当场比赛规则为准；开赛时间不代表已核实的报名截止。
Source: Codeforces · Official source · Last checked: Oct 30, 2026, 10:27 AM (Asia/Shanghai)
Current source coverage: Currently covers the official Databricks and Scale AI job boards, mainly internship and graduate roles overseas. Check location, graduation-year and work-authorization requirements in each original listing.
No new content in this edition
The checked sources have no new, unselected content. You can read an earlier edition.