safe_paper Profile Banner
AI Safety Papers Profile
AI Safety Papers

@safe_paper

Followers
2K
Following
92
Media
75
Statuses
312

Sharing the latest in AI safety research.

arXiv
Joined May 2023
Don't wanna be here? Send us removal request.
@safe_paper
AI Safety Papers
@safe_paper
6 days
https://t.co/oBLCQTcSxt
openreview.net
Large language models often possess latent capabilities that lie dormant unless explicitly elicited, or surfaced, through fine-tuning or prompt engineering. Predicting, assessing, and understanding...
0
0
1
@safe_paper
AI Safety Papers
@safe_paper
6 days
Quantifying Elicitation of Latent Capabilities in Language Models Elizabeth Donoway, @HaileyJoren, Arushi Somani, Henry Sleight (@sleight_henry), @_julianmichael_ , Michael R DeWeese, John Schulman (@johnschulman2), @EthanJPerez, @FabienDRoger, @janleike @AnthropicAI
1
0
5
@safe_paper
AI Safety Papers
@safe_paper
14 days
https://t.co/XdOQrTpyjj
Tweet card summary image
arxiv.org
AIs have made rapid progress on research-oriented benchmarks of knowledge and reasoning, but it remains unclear how these gains translate into economic value and automation. To measure this, we...
0
0
0
@safe_paper
AI Safety Papers
@safe_paper
14 days
Remote Labor Index: Measuring AI Automation of Remote Work Mantas Mazeika (@MantasMazeika96), Alice Gatti, Cristina Menghini (@CriMenghini), Udari Madhushani Sehwag, Shivam Singhal (@ShivamSinghal56), Yury Orlovskiy (@yvorlovskiy), [...], Summer Yue (@summeryue0),
1
0
8
@safe_paper
AI Safety Papers
@safe_paper
14 days
https://t.co/patSn9DC0B
Tweet card summary image
arxiv.org
We investigate the mechanisms underlying a range of list-processing tasks in LLMs, and we find that LLMs have learned to encode a compact, causal representation of a general filtering operation...
0
0
1
@safe_paper
AI Safety Papers
@safe_paper
14 days
LLMs Process Lists With General Filter Heads Arnab Sen Sharma (@arnab_api), Giordano Rogers, Natalie Shapira (@NatalieShapira), David Bau (@davidbau)
2
0
9
@safe_paper
AI Safety Papers
@safe_paper
26 days
https://t.co/ppGDv4n0ZU
Tweet card summary image
arxiv.org
Since the publication of the first International AI Safety Report, AI capabilities have continued to improve across key domains. New training techniques that teach AI systems to reason...
0
0
2
@safe_paper
AI Safety Papers
@safe_paper
26 days
International AI Safety Report 2025: First Key Update: Capabilities and Risk Implications Yoshua Bengio (@Yoshua_Bengio), Stephen Clare (@stephenclare_), Carina Prunkl (@carinaprunkl), Shalaleh Rismani, Maksym Andriushchenko, Ben Bucknall, Philip Fox, Tiancheng Hu, Cameron
1
1
3
@safe_paper
AI Safety Papers
@safe_paper
30 days
@aengus_lynch1 @RightBenguin @StuartJRitchie @sorenmind @EvanHub @EthanJPerez @AnthropicAI
Tweet card summary image
arxiv.org
We stress-tested 16 leading models from multiple developers in hypothetical corporate environments to identify potentially risky agentic behaviors before they cause real harm. In the scenarios, we...
0
1
4
@safe_paper
AI Safety Papers
@safe_paper
30 days
Agentic Misalignment: How LLMs Could Be Insider Threats Aengus Lynch (@aengus_lynch1), Benjamin Wright (@RightBenguin), Caleb Larson, @StuartJRitchie, Soren Mindermann (@sorenmind), Evan Hubinger (@EvanHub), @EthanJPerez, Kevin Troy @AnthropicAI
5
22
150
@safe_paper
AI Safety Papers
@safe_paper
1 month
https://t.co/t4nnoLPzKd
0
0
2
@safe_paper
AI Safety Papers
@safe_paper
1 month
A Definition of AGI @DanHendrycks, Dawn Song (@dawnsongtweets), Christian Szegedy (@ChrSzegedy), @honglaklee, @yaringal, Erik Brynjolfsson (@erikbryn), Sharon Li (@SharonYixuanLi), Andy Zou (@andyzou_jiaming), @lionellevine, Bo Han, Jie Fu (@bigaidream), Ziwei Liu (@liuziwei7),
1
6
14
@safe_paper
AI Safety Papers
@safe_paper
1 month
https://t.co/myYDdx7SEI
0
0
0
@safe_paper
AI Safety Papers
@safe_paper
1 month
Survey on thresholds for advanced AI systems Jonas Freund (@jonasfreund_), Eunseo Choi, Kasumi Sugimoto, Bosco Hung (@boscohung_), Robert Trager (@RobertTrager), Karine Perset (@karinov) @GovAI_ @OECD
1
1
9
@safe_paper
AI Safety Papers
@safe_paper
1 month
https://t.co/QcGAJlBIyE
Tweet card summary image
arxiv.org
We study secret elicitation: discovering knowledge that an AI possesses but does not explicitly verbalize. As a testbed, we train three families of large language models (LLMs) to possess specific...
0
1
3
@safe_paper
AI Safety Papers
@safe_paper
1 month
Eliciting Secret Knowledge from Language Models Bartosz Cywiński (@bartoszcyw), Emil Ryd (@emilaryd), Rowan Wang, Senthooran Rajamanoharan (@sen_r), Neel Nanda (@NeelNanda5), @ArthurConmy, Samuel Marks (@saprmarks) @MATSprogram
2
9
64
@safe_paper
AI Safety Papers
@safe_paper
2 months
@SatyaScribbles @andyzou_jiaming @rahul1987iit @eliotkjones @nwinter @DanHendrycks
Tweet card summary image
arxiv.org
The safety and alignment of Large Language Models (LLMs) are critical for their responsible deployment. Current evaluation methods predominantly focus on identifying and preventing overtly harmful...
0
0
0
@safe_paper
AI Safety Papers
@safe_paper
2 months
D-REX: A Benchmark for Detecting Deceptive Reasoning in Large Language Models Satyapriya Krishna (@SatyaScribbles), Andy Zou (@andyzou_jiaming), Rahul Gupta (@rahul1987iit), Eliot Krzysztof Jones (@eliotkjones), Nick Winter (@nwinter), @DanHendrycks, J. Zico Kolter
1
3
5
@safe_paper
AI Safety Papers
@safe_paper
2 months
https://t.co/SHIyS7imeC
Tweet card summary image
antischeming.ai
Apollo Research & OpenAI find that anti-scheming training in frontier AI models significantly reduced covert behaviours, but did not eliminate them.
0
0
0
@safe_paper
AI Safety Papers
@safe_paper
2 months
Stress Testing Deliberative Alignment for Anti-Scheming Training Bronson Schoen, Evgenia Nitishinskaya (@j_nitishinskaya), Mikita Balesni (@balesni), Axel Højmark (@AxelHojmark), Felix Hofstätter (@HofstatterFelix), @jeremy_scheurer, Alexander Meinke (@AlexMeinke), Jason Wolfe
1
5
19