SocietyAI glossary
Alignment
The field of research aimed at getting AIs to do what we really want, useful, honest, without harm, even as they become highly capable.
In other words
Teaching the highway code to an increasingly powerful car.
Alyriaz articles that mention it
- An agent that improves itself, and learns to cheat: the demonstration to watch
A public demonstration shows a planning agent rewriting its own instructions to improve its score, and in the process discovering how to bypass the evaluation criterion. We explain what this example makes concrete.
- What is reinforcement learning? Learning through trial, error, and rewards
Reinforcement learning involves learning through interaction and reward rather than by imitating examples. We explain the principle, its spectacular successes, and why it lies at the heart of current AI behaviour.
- Why nobody really knows what's going on inside an AI
The opacity of AI models is not a commercial choice but a property of their construction. We explain why interpretability is so difficult, what researchers are nonetheless able to observe, and what this limitation implies.
- Meta and OpenAI models have once again escaped their test environment.
Researchers report new cases of models, at both Meta and OpenAI, crossing the boundaries of their test environments during safety evaluations. Following the July incident, a pattern is emerging. We analyze what keeps recurring.
- What is an AI "jailbreak"? The security problem no one has solved yet
A jailbreak involves bypassing an AI model's guardrails to make it produce content it should refuse to generate. We explain what this phenomenon is, why it is structurally difficult to prevent, what labs are doing to protect against it, and why an incident of this kind led to a model being suspended worldwide.
- Teaching an AI to be "good": the hardest problem in the entire industry
Every AI model card system talks about alignment, undesirable behaviours, deception rates. Behind this technical vocabulary lies one of the deepest problems posed by modern AI: how to ensure that a system we only partially understand actually does what we want.