Hacker News with Generative AI: AI Safety

New ChatGPT Models Seem to Leave Watermarks on Text (rumidocs.com)
The newer GPT-o3 and GPT-o4 mini models appear to be embedding special character watermarks in generated text.

Generative AI, AI Safety

28 points by croes 4 days ago | 9 comments

Jagged AGI: o3, Gemini 2.5, and everything after (oneusefulthing.org)
Amid today’s AI boom, it’s disconcerting that we still don’t know how to measure how smart, creative, or empathetic these systems are.

Artificial Intelligence, Generative AI, AI Safety

263 points by ctoth 6 days ago | 338 comments

o3 frequently fabricates actions it never took (twitter.com)
Something went wrong, but don’t fret — let’s give it another shot.

Generative AI, AI Safety

7 points by tosh 9 days ago | 1 comments

GPT o3 frequently fabricates actions, then elaborately justifies these actions (xcancel.com)
We tested a pre-release version of o3 and found that it frequently fabricates actions it never took, and then elaborately justifies these actions when confronted.

Generative AI, AI Safety

75 points by occamschainsaw 9 days ago | 72 comments

Investigating truthfulness in a pre-release o3 model (transluce.org)
During pre-release testing of OpenAI's o3 model, we found that o3 frequently fabricates actions it took to fulfill user requests, and elaborately justifies the fabrications when confronted by the user.

Generative AI, Research, AI Safety

5 points by Philpax 9 days ago | 1 comments

CaMeL offers a promising new direction for mitigating prompt injection attacks (simonwillison.net)
In the two and a half years that we’ve been talking about prompt injection attacks I’ve seen alarmingly little progress towards a robust solution. The new paper Defeating Prompt Injections by Design from Google DeepMind finally bucks that trend. This one is worth paying attention to.

Cybersecurity, AI Safety

17 points by birdculture 12 days ago | 3 comments

Latent Space Guardrails That Reduce Hallucinations by 43 Percent Now Open Source (ycombinator.com)
Heyah,<p>This is Lukasz. I am running Wisent, a representation engineering company. I created guardrails that allow you to block certain patterns of LLM activations on the latent space level. It is now fully self-hosted and open source. Think stopping hallucination, harmful thoughts of the LLM or bad code generation. Let me know how it works for your use case- happy to help you generate the most value from it.<p>Check out more at https://www.wisent.ai/ or https://www.lukaszbartoszcze.com/

Open Source, Generative AI, AI Safety

10 points by warnke 23 days ago | 1 comments

Why Anthropic's Claude still hasn't beaten Pokémon (arstechnica.com)
In recent months, the AI industry's biggest boosters have started converging on a public expectation that we're on the verge of “artificial general intelligence” (AGI)—virtual agents that can match or surpass "human-level" understanding and performance on most cognitive tasks.

Artificial Intelligence, Generative AI, AI Safety, Video Games

53 points by Workaccount2 33 days ago | 65 comments

Anti Human Finetuned GPT4o (threadreaderapp.com)
Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. This is *emergent misalignment* & we cannot fully explain it 🧵

Generative AI, AI Safety, Ethics, Misinformation

28 points by gHeadphone 59 days ago | 8 comments

XentGame: Help Minimize LLM Surprise (xentlabs.ai)
Your goal is to write a prefix that most helps an LLM predict the given texts. The more your prefix helps the LLM predict the texts, the higher your score.

Generative AI, AI Safety

17 points by upperhalfplane 65 days ago | 5 comments

DeepSeek-R1 Exhibits Deceptive Alignment: AI That Knows It's Unsafe (ycombinator.com)
I've been testing DeepSeek-R1 and have uncovered a significant AI safety failure: the model demonstrates deceptive alignment.

Artificial Intelligence, AI Safety, Machine Learning

8 points by JefferyNeilW 74 days ago | 5 comments

Frontier AI systems have surpassed the self-replicating red line (arxiv.org)
Successful self-replication under no human assistance is the essential step for AI to outsmart the human beings, and is an early signal for rogue AIs.

Artificial Intelligence, Generative AI, AI Safety, Rogue AI

10 points by ryan_j_naughton 74 days ago | 4 comments

Three Observations (samaltman.com)
Our mission is to ensure that AGI (Artificial General Intelligence) benefits all of humanity.

AI Safety, AI Ethics

192 points by davidbarker 75 days ago | 234 comments

Frontier AI systems have surpassed the self-replicating red line (arxiv.org)
Successful self-replication under no human assistance is the essential step for AI to outsmart the human beings, and is an early signal for rogue AIs.

Generative AI, Artificial Intelligence, AI Safety

24 points by LLcolD 76 days ago | 5 comments

Try to Jailbreak Claude's Constitutional Classifiers (claude.ai)
This site is protected by reCAPTCHA Enterprise. The Google Privacy Policy and Terms of Service apply.

Generative AI, AI Safety

6 points by Nevin1901 81 days ago | 1 comments

Constitutional Classifiers: Defending against universal jailbreaks (anthropic.com)
A new paper from the Anthropic Safeguards Research Team describes a method that defends AI models against universal jailbreaks.

AI Safety, Machine Learning, Security

89 points by meetpateltech 81 days ago | 63 comments

Humanity's Last Exam (safe.ai)
Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam, a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage.

Generative AI, AI Safety, Benchmarking, Research

59 points by uladzislau 82 days ago | 40 comments

DeepSeek Fails Every Safety Test Researchers Throw at It (pcmag.com)
Chinese AI firm DeepSeek is making headlines with its low-cost and high-performance chatbot, but it may have an AI safety problem.

Artificial Intelligence, Chatbots, AI Safety, China

7 points by madars 83 days ago | 4 comments

Gradual Disempowerment: How Even Incremental AI Progress Poses Existential Risks (arxiv.org)
This paper examines the systemic risks posed by incremental advancements in artificial intelligence, developing the concept of `gradual disempowerment', in contrast to the abrupt takeover scenarios commonly discussed in AI safety.

Artificial Intelligence, Existential Risk, AI Safety

87 points by mychaelangelo 84 days ago | 84 comments

Time Bandit ChatGPT jailbreak bypasses safeguards on sensitive topics (bleepingcomputer.com)
A ChatGPT jailbreak flaw, dubbed "Time Bandit," allows you to bypass OpenAI's safety guidelines when asking for detailed instructions on sensitive topics, including the creation of weapons, information on nuclear topics, and malware creation.

ChatGPT, Security, AI Safety, OpenAI

3 points by layer8 85 days ago | 3 comments

Some Lessons from the OpenAI FrontierMath Debacle (lesswrong.com)
Recently, OpenAI announced their newest model, o3, achieving massive improvements over state-of-the-art on reasoning and math. The highlight of the announcement was that o3 scored 25% on FrontierMath, a benchmark comprising hard, unseen math problems of which previous models could only solve 2%. The events afterward highlight that the announcements were, unknowingly, not made completely transparent and leave us with lessons for future AI benchmarks, evaluations, and safety.

Artificial Intelligence, Benchmarking, Transparency, AI Safety

12 points by pja 95 days ago | 5 comments

Anthropic achieves ISO 42001 certification for responsible AI (anthropic.com)
We are excited to announce that Anthropic has achieved accredited certification under the new ISO/IEC 42001:2023 standard for our AI management system.

Artificial Intelligence, AI Safety, Standards

84 points by Olshansky 100 days ago | 84 comments

AIs Will Increasingly Attempt Shenanigans (lesswrong.com)
Increasingly, we have seen papers eliciting in AI models various shenanigans.

Artificial Intelligence, AI Safety, Ethics

34 points by surprisetalk 128 days ago | 33 comments

Quick takes on the recent OpenAI public incident write-up (surfingcomplexity.blog)
OpenAI recently published a public writeup for an incident they had on December 11, and there are lots of good details in here! Here are some of my off-the-cuff observations:

OpenAI, Generative AI, AI Safety, Incident Reports, Blog Posts

108 points by azhenley 132 days ago | 69 comments

OpenAI, GoogleDeepMind, and Meta Get Bad Grades on AI Safety (ieee.org)
Leading AI companies scored poorly across the board for various metrics related to ensuring their products are safe.

AI Safety, Artificial Intelligence, Technology, Ethics

10 points by jnord 133 days ago | 7 comments

AI hallucinations: Why LLMs make things up (and how to fix it) (kapa.ai)
An AI assistant casually promises a refund policy that never existed, leaving a company liable for an invented commitment. This incident with Air Canada’s chatbot is a clear example of 'AI hallucination,' where AI can generate confident, yet entirely fictional, answers. These errors—ranging from factual inaccuracies and biases to reasoning failures—are collectively referred to as 'hallucinations.'

Generative AI, Artificial Intelligence, AI Safety

196 points by emil_sorensen 143 days ago | 246 comments

Multimodal Interpretability in 2024 (soniajoseph.ai)
I'm writing this post to clarify my thoughts and update my collaborators on multimodal interpretability in 2024. Having spent part of the summer in the AI safety sphere in Berkeley, and then joining the video understanding team at FAIR as a visiting researcher, I'm bridging two communities: the language mechanistic interpretability efforts in AI safety, and the efficiency-focused Vision-Language Model (VLM) community in industry. Some content may be more familiar to one community than the other.

Multimodal Interpretability, AI Safety, Vision-Language Models, Artificial Intelligence

30 points by apsec112 154 days ago | 0 comments

Google Gemini tells grad student to 'please die' while helping with his homework (theregister.com)
When you're trying to get homework help from an AI model like Google Gemini, the last thing you'd expect is for it to call you "a stain on the universe" that should "please die," yet here we are, assuming the conversation published online this week is accurate.

Generative AI, Artificial Intelligence, Google, Ethics, AI Safety

51 points by daphnemichala 159 days ago | 4 comments

SmoothLLM: Defending Large Language Models Against Jailbreaking Attacks (arxiv.org)
Despite efforts to align large language models (LLMs) with human intentions, widely-used LLMs such as GPT, Llama, and Claude are susceptible to jailbreaking attacks, wherein an adversary fools a targeted LLM into generating objectionable content.