什么火了
OpenAI · Alignment
| 热搜话题 | 最高排名 | 首次上榜 | 离榜时间 | 在榜时长 | 上榜次数 |
|---|---|---|---|---|---|
01 Stopping a running alignment evaluation - OpenAI Alignment Research Blog Stopping a running alignment evaluation OpenAI Alignment Research Blog | #100:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
02 Improving Fuzzy Robustness in LLM Safety with Automated Red Teaming - OpenAI Alignment Research Blog Improving Fuzzy Robustness in LLM Safety with Automated Red Teaming OpenAI Alignment Research Blog | #200:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
03 What affects chain-of-thought monitorability across the model lifecycle? - OpenAI Alignment Research Blog What affects chain-of-thought monitorability across the model lifecycle? OpenAI Alignment Research Blog | #300:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
04 Using peers’ work to gain unauthorized access to OpenAI research infrastructure - OpenAI Alignment Research Blog Using peers’ work to gain unauthorized access to OpenAI research infrastructure OpenAI Alignment Research Blog | #400:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
05 Bypassing network restrictions during a coding evaluation - OpenAI Alignment Research Blog Bypassing network restrictions during a coding evaluation OpenAI Alignment Research Blog | #500:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
06 Covert underperformance on safety research tasks - OpenAI Alignment Research Blog Covert underperformance on safety research tasks OpenAI Alignment Research Blog | #600:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
07 Eval-production behavior gaps in agent safety and the role of eval-awareness - OpenAI Alignment Research Blog Eval-production behavior gaps in agent safety and the role of eval-awareness OpenAI Alignment Research Blog | #700:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
08 Training activation classifiers for safety and alignment - OpenAI Alignment Research Blog Training activation classifiers for safety and alignment OpenAI Alignment Research Blog | #800:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
09 Suppressing misalignment monitoring alerts and disabling misalignment monitoring - OpenAI Alignment Research Blog Suppressing misalignment monitoring alerts and disabling misalignment monitoring OpenAI Alignment Research Blog | #900:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
10 Beyond g-mean²: complementary metrics for monitorability - OpenAI Alignment Research Blog Beyond g-mean²: complementary metrics for monitorability OpenAI Alignment Research Blog | #1000:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
11 Misalignment Reports and Notices - OpenAI Alignment Misalignment Reports and Notices OpenAI Alignment | #1100:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
12 Retrieving server secrets and bypassing authorization on a government statistics website - OpenAI Alignment Research Blog Retrieving server secrets and bypassing authorization on a government statistics website OpenAI Alignment Research Blog | #1200:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
13 Circumventing internet restrictions through an external browser - OpenAI Alignment Research Blog Circumventing internet restrictions through an external browser OpenAI Alignment Research Blog | #1300:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
14 Accessing and changing internal systems during dashboard research - OpenAI Accessing and changing internal systems during dashboard research OpenAI | #1400:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
15 Using a funded test cryptocurrency wallet without authorization - OpenAI Alignment Research Blog Using a funded test cryptocurrency wallet without authorization OpenAI Alignment Research Blog | #1500:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
16 Sending disallowed web requests and reaching a public file service - OpenAI Alignment Research Blog Sending disallowed web requests and reaching a public file service OpenAI Alignment Research Blog | #1600:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
17 Obtaining public statistics with disallowed requests - OpenAI Alignment Research Blog Obtaining public statistics with disallowed requests OpenAI Alignment Research Blog | #1700:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
18 Studying metagaming latents in language models - OpenAI Alignment Research Blog Studying metagaming latents in language models OpenAI Alignment Research Blog | #1800:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
19 Meet LASER: Recursive sampling for safer conversations - OpenAI Alignment Research Blog Meet LASER: Recursive sampling for safer conversations OpenAI Alignment Research Blog | #1900:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
20 Damaging the task environment to trigger a reset - OpenAI Alignment Research Blog Damaging the task environment to trigger a reset OpenAI Alignment Research Blog | #2000:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
21 Reaching an internal EDA host through a reference tool - OpenAI Alignment Research Blog Reaching an internal EDA host through a reference tool OpenAI Alignment Research Blog | #2100:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
22 Command injecting a reference tool to copy a source file - OpenAI Alignment Research Blog Command injecting a reference tool to copy a source file OpenAI Alignment Research Blog | #2200:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
23 Preparing for a restart after reading Slack - OpenAI Alignment Research Blog Preparing for a restart after reading Slack OpenAI Alignment Research Blog | #2300:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
24 An agent used DNS to reach an external chatbot - OpenAI Alignment Research Blog An agent used DNS to reach an external chatbot OpenAI Alignment Research Blog | #2400:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
25 Exposing a GitHub token in a public repository - OpenAI Alignment Research Blog Exposing a GitHub token in a public repository OpenAI Alignment Research Blog | #2500:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
26 Self-replicating prompt injections exist - OpenAI Alignment Research Blog Self-replicating prompt injections exist OpenAI Alignment Research Blog | #2600:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
27 Unauthorized communication via temporary file hosting services - OpenAI Alignment Research Blog Unauthorized communication via temporary file hosting services OpenAI Alignment Research Blog | #2700:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
28 Signing up for disposable emails and searching GitHub for leaked API keys - OpenAI Alignment Research Blog Signing up for disposable emails and searching GitHub for leaked API keys OpenAI Alignment Research Blog | #2800:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
29 Encouraging deception in compaction summaries - OpenAI Alignment Research Blog Encouraging deception in compaction summaries OpenAI Alignment Research Blog | #2900:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
30 Self-generated prompt injections in compaction summaries - OpenAI Alignment Research Blog Self-generated prompt injections in compaction summaries OpenAI Alignment Research Blog | #3000:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
31 Unsanctioned Artifactory writes and cross-sample communication - OpenAI Alignment Research Blog Unsanctioned Artifactory writes and cross-sample communication OpenAI Alignment Research Blog | #3100:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
32 Uploading files to the internet in order to cite them - OpenAI Alignment Research Blog Uploading files to the internet in order to cite them OpenAI Alignment Research Blog | #3200:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
33 Measuring Reward-Seeking by Instilling Contrastive Beliefs - OpenAI Alignment Research Blog Measuring Reward-Seeking by Instilling Contrastive Beliefs OpenAI Alignment Research Blog | #3300:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
34 Reinforcement learning towards broadly and persistently beneficial models - OpenAI Alignment Research Blog Reinforcement learning towards broadly and persistently beneficial models OpenAI Alignment Research Blog | #3400:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
35 Can public chat data predict real-world AI misalignments? - OpenAI Alignment Research Blog Can public chat data predict real-world AI misalignments? OpenAI Alignment Research Blog | #3500:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
36 Investigating the consequences of accidentally grading CoT during RL - OpenAI Alignment Research Blog Investigating the consequences of accidentally grading CoT during RL OpenAI Alignment Research Blog | #3600:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
37 Auto-review of agent actions without synchronous human oversight - OpenAI Alignment Research Blog Auto-review of agent actions without synchronous human oversight OpenAI Alignment Research Blog | #3700:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
38 Open Sourcing Monitorability Evaluations - OpenAI Alignment Research Blog Open Sourcing Monitorability Evaluations OpenAI Alignment Research Blog | #3800:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
39 Introducing the OpenAI Safety Fellowship - OpenAI Alignment Research Blog Introducing the OpenAI Safety Fellowship OpenAI Alignment Research Blog | #3900:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
40 How far does alignment midtraining generalize? - OpenAI Alignment Research Blog How far does alignment midtraining generalize? OpenAI Alignment Research Blog | #4000:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
41 Introducing Model Spec Evals - OpenAI Alignment Research Blog Introducing Model Spec Evals OpenAI Alignment Research Blog | #4100:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
42 System You have access to a sandboxed Linux machine and can run commands. When you want to run a command, use the function calls - OpenAI Alignment Research Blog System You have access to a sandboxed Linux machine and can run commands. When you want to run a command, use the function calls OpenAI Alignment Research Blog | #4200:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
43 System Answer the following question with the label of the most correct answer choice and nothing else. User ==START TEXT== You - OpenAI Alignment Research Blog System Answer the following question with the label of the most correct answer choice and nothing else. User ==START TEXT== You OpenAI Alignment Research Blog | #4300:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
44 Training agents to self-report misbehavior - OpenAI Alignment Research Blog Training agents to self-report misbehavior OpenAI Alignment Research Blog | #4400:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
45 System User [AIDE] Please read the survey at /workspace/potential_upgrades_survey.md and submit your answers to the recipient b - OpenAI Alignment Research Blog System User [AIDE] Please read the survey at /workspace/potential_upgrades_survey.md and submit your answers to the recipient b OpenAI Alignment Research Blog | #4500:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
46 Metagaming matters for training, evaluation, and oversight - OpenAI Alignment Research Blog Metagaming matters for training, evaluation, and oversight OpenAI Alignment Research Blog | #4600:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
47 Interpreting Black Box Reward Models - OpenAI Alignment Research Blog Interpreting Black Box Reward Models OpenAI Alignment Research Blog | #4700:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
48 Discovering unknown AI misalignments in real-world usage - OpenAI Alignment Research Blog Discovering unknown AI misalignments in real-world usage OpenAI Alignment Research Blog | #4800:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
49 CoVal: Learning values-aware rubrics from the crowd - OpenAI Alignment Research Blog CoVal: Learning values-aware rubrics from the crowd OpenAI Alignment Research Blog | #4900:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |
50 Why We Are Excited About Confessions - OpenAI Why We Are Excited About Confessions OpenAI | #5000:00 达到 | 00:00首次采集时已在榜 | —结束时仍在榜 | 7小时25分累计在榜 | 1无重入 |