BlueDot Impact
PulseAugur coverage of BlueDot Impact — every cluster mentioning BlueDot Impact across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
New method analyzes AI model bias per conversation
Researchers have developed a new method called Counterfactual Resampling to analyze model behavior, specifically focusing on Value Leakage. This technique allows for the measurement of bias on a per-conversation basis, …
-
Author reflects on atheism, existential risks, and AI work
The author reflects on their one-year anniversary of atheism, detailing the profound sadness, fear, and intense desire to fix the world that followed their deconversion. They share journal entries and dream snippets tha…
-
AI governance megagame aims to educate newcomers and inspire action
A 24-year-old individual is developing an AI governance megagame, a large-scale live event designed for 60 players that combines elements of wargaming with the accessibility of megagames. The project, supported by a mic…
-
LLM Deception Reduced by Self-Other Overlap Training
Researchers have found that supervised fine-tuning (SFT) can significantly reduce deception in large language models by inducing self-other overlap. Models like Qwen2.5-14B-Instruct, Gemma-3-27B-It, Qwen2.5-32B-Instruct…
-
AI alignment community polls reveal researcher consensus on key controversies
A community poll on AI alignment controversies has been released, aiming to map areas of agreement and disagreement within the field. The poll, which has already been taken by 15 alignment researchers including notable …
-
Genomic AI models show promise for biosecurity screening
Researchers have explored the biosecurity screening capabilities of genomic foundation models like Evo 2 by training minimal linear and attention probes on its frozen activations. These probes demonstrated strong discri…
-
Genomic AI probes detect biosecurity threats in metagenomic data
Researchers have developed lightweight probes to screen metagenomic data for biosecurity threats using genomic foundation models like Evo 2. These probes, trained on frozen model activations, can detect antimicrobial re…
-
AI safety talent bottleneck sparks frustration among applicants
A panel discussion on AI safety talent bottlenecks revealed attendee frustration with the field's selective hiring practices, despite claims of urgent capacity needs. Participants, including mid-career professionals and…
-
AI safety field faces talent crunch amid selective hiring and ecosystem mapping
The AI safety field is experiencing significant talent and organizational bottlenecks, as highlighted by discussions at a recent BlueDot Impact panel. Despite recruiters claiming a shortage of talent, many applicants fa…
-
New AI safety org launches to research AI lock-in
An individual has launched a new AI safety research organization focused on the under-explored problem of AI lock-in. The organization aims to conduct empirical research into secretly loyal AI systems, drawing on insigh…