Top Language Models Proven Most Resistant to Russian Disinformation
benchmarks
| Source: Mastodon | Original article
Large language models resist Russian propaganda. Top models defy foreign disinformation.
As we reported on June 5, concerns have been growing about large language models (LLMs) spreading Russian propaganda. Now, a benchmark by the Estonian government has identified the LLMs most resistant to Russian disinformation. The study reveals that recent models have improved significantly in combating "strategic narratives" promoted by Russia.
This development matters because state governments are increasingly worried about the potential for LLMs to spread harmful propaganda. The Estonian benchmark shows that some models are more effective than others in resisting Russian disinformation, with newer models outperforming their predecessors. For instance, the highest-rated model from 2024, Claude 3.5 Haiku, scored lower than many 2026 models in resisting Russian propaganda.
Looking ahead, it will be crucial to monitor how these findings influence the development of LLMs and their potential applications. As Russia continues to exploit new methods to spread disinformation, including LLM grooming, the ability of these models to resist propaganda will be essential in maintaining the integrity of online information. The Estonian government's benchmark provides valuable insights, and further research is needed to ensure that LLMs can effectively counter Russian disinformation efforts.
Sources
Back to AIPULSEN