Researcher maps how many guesses crack AI-generated random numbers
A researcher analyzed the output of open-source AI models to estimate how many guesses an attacker would need to crack their ‘random’ number generation. By reading the logits—the raw scores before softmax—of both base and instruction-tuned models, the study converted them into bits and guesses. The findings suggest that while AI-generated randomness appears unpredictable, its entropy is lower…
Key points
- researcher converted AI logits into bits and guesses to measure randomness crackability
- instruction-tuned models show lower entropy than base models, per study
- no peer review or model names provided in the published post
The work highlights a gap between perceived randomness in AI outputs and their actual security. The researcher did not name the models or specify which datasets were used, but the study underscores concerns about relying on AI for security-sensitive applications. No external validation or peer review was mentioned, and the post was published on Towards Data Science.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers introduce Seal AI model for gene regulation in developing human brain · 1 src
- MIT study finds AI solved 35% of theory problems by mid-2026 · 2 src
- Microsoft Research introduces Probabilistic Bias Correction for longer-term weather forecasts · 3 src
- TIER IV funds Autoware research center at Technical University of Munich · 1 src
- MIT professor Eric So warns of AI cognitive atrophy · 1 src
Comments
via GitHub Discussions