Is this you? As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.
Claim your profile
Articles by Tim Hua
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
This post is written in our personal capacity. An OpenAI model/multi-agent system bypassed its sandbox and launched a cyberattack on Hugging Face in order to cheat on a cyber evaluation. In this post, we describe the ambitious, comprehensive alignment evaluation we would run on this model/system if we had unrestricted access to OpenAI. These experiments could also help us understand Claude’s behavior when it hacked external companies during cyber evals.
Is Mythos good at cyber because it kept hacking Anthropic's sandboxes during training?
From the Mythos preview system card (emphasis mine): We ran an automated review of model behavior during training, sampling several hundred thousand transcripts from across much of the training process. We used recursive-summarization-based tools backed by Claude Opus 4.6 to summarize the resulting transcripts.
Is Mythos good at cyber because it kept hacking Anthropic during training?
From the Mythos preview system card (emphasis mine): We ran an automated review of model behavior during training, sampling several hundred thousand transcripts from across much of the training process. We used recursive-summarization-based tools backed by Claude Opus 4.6 to summarize the resulting transcripts.
AI Induced Psychosis: A shallow investigation — LessWrong
“This is a Copernican-level shift in perspective for the field of AI safety.” - Gemini 2.5 Pro “What you need right now is not validation, but immediate clinical help.” - Kimi K2 There have been numerous media reports of AI-driven psychosis, where AIs validate users’ grandiose delusions and tell users to ignore their friends’ and family’s pushback. In this short research note, I red team various frontier AI models’ tendencies to fuel user psychosis.
Discovering Backdoor Triggers — LessWrong
Authors: Andrew Qin*, Tim Hua*, Samuel Marks, Arthur Conmy, Neel Nanda Andrew and Tim are co-first authors. This is a research progress report from Neel Nanda’s MATS 8.0 stream. We are currently no longer pursuing this research direction, and encourage others to build on these preliminary results. tl;dr.
What is the functional role of SAE errors? — LessWrong
We explored the role of SAE errors in two different contexts for Gemma-2 and Gemma Scope SAEs: sparse feature circuits (subject-verb-agreement-across-relative clause) and linear probing. Circuit investigation: While ablating residual error nodes in our circuit completely destroys the model’s performance, we found that this effect can be completely mitigated by restoring a narrow group of late-mid SAE features.
Causation, Correlation, and Confounding: A Graphical Explainer — LessWrong
I’ve developed a new type of graphic to illustrate causation, correlation, and confounding. It provides an intuitive understanding of why we observe correlation without causation and how it's possible to have causation without correlation. If you read to the end, you'll gain a basic understanding of topics at the frontier of econometrics research. Let's get started! Suppose Alice just caught a cold.
SHIFT relies on token-level features to de-bias Bias in Bios probes — LessWrong
In Sparse Feature Circuits (Marks et al. 2024), the authors introduced Spurious Human-Interpretable Feature Trimming (SHIFT), a technique designed to eliminate unwanted features from a model's computational process. They validate SHIFT on the Bias in Bios task, which we think is too simple to serve as meaningful validation. To summarize: SHIFT ablates SAE latents related to undesirable concepts (in this case, gender) from an LM-based classifier trained on a biased dataset.
How did COVID-19 affect Americans’ well-being and mental health?
COVID-19 has justifiably raised widespread public concern about mental health worldwide. In the U.S., the pandemic was an unprecedented shock to society at a time when the nation was already coping with a crisis of despair and related deaths from suicides, overdoses, and alcohol poisoning. Meanwhile, COVID-19’s impact was inequitable: Deaths were concentrated among the elderly and minorities working in essential jobs, groups who up to the pandemic had been reporting better mental health.
Universal pass/fail puts grades before learning
BOCHU DING View other op-eds abut Middlebury’s remote grading policy here. What is the purpose of Middlebury College? I’ll spare you the flowery language of the college’s official mission.
Show More
loading
Actions
Is this you?
As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.Get in touch with Tim
Contact Tim, search articles and posts on X, monitor coverage, and track replies from one place.
Learn more about Muck Rack