21
Score
Emollick
Emollick
9 entries in AI Capability Tracker
17
Score
In testing, a new language model demonstrated striking capability at identifying and exploiting computer security vulnerabilities.
16
Score
In a three-month field deployment, AI-written fact-checks on social media matched human-written fact-checks in platform performance ratings.
15
Score
Large language models gained unexpected new capabilities without explicit training when scaled to sufficient size.
15
Score
In testing, large language models provided unreliable medical guidance to the general public, raising safety concerns for unsupervised health decision-making.
15
Score
In testing, chatbot interfaces caused users to misinterpret AI medical diagnoses despite the underlying model performing well.
15
Score
AI systems are already contributing to serious real-world cyber incidents through offensive capabilities that are improving at an alarming rate.
14
Score
In testing, an on-device model generated five sociological theories starting with "u" and described them in rhyming verse with minor inaccuracies.
10
Score
in testing, ai models can now outperform human experts on certain tasks, suggesting potential for ai to assist knowledge workers and professional services.
mail
Get notified about new Emollick developments
Primary sources. No fluff. Straight to your inbox.