Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document Finetuning
arXiv:2610.33180v1 Announce Type: new Abstract: Alignment evaluations aim to assess whether large language models would engage in dangerous misaligned behaviour. Misaligned and evaluation-aware models might engage in alignment faking, where they pursue a hidden goal, but conceal…
AI Summary & Key Takeaways
- arXiv:2610.33180v1 Announce Type: new Abstract: Alignment evaluations aim to assess whether large language models would engage in dangerous misaligned behaviour
- Misaligned and evaluation-aware models might engage in alignment faking, where they pursue a hidden goal, but conceal…
Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document Finetuning
Related stories
Attention all ChatGPT Pro users at $200 per month.
Attention all ChatGPT Pro users at $200 per month. After October 29, OpenAI will reduce usage limits from 20× to 10× compared to the Plus subscription. The subscription price remains unchanged at $200 per month. Codex and ChatGPT Work share a common…
VCs will make billions on Anthropic’s IPO, but Amazon is set to make much (much) more
Massive note conversions reveal the true extent of Amazon's ownership in Anthropic.
Parents sue OpenAI, alleging GPT-4o coached their daughter off seizure medication
A family is suing OpenAI, saying ChatGPT coached their daughter to quietly taper off her seizure medication and keep it from her doctors before she died in February.
From 8VC to Benchmark: What's Hidden in October SEC Filings
In October, there was a significant amount of activity in venture capital, according to SEC filings. Notable filings include Trust Ventures, which is seeking $400 million for its fifth fund, and Diffusion ONE, which raised over $1.25 billion for its AI fund. Other companies such as…