Comment by delifue
5 days ago
An interesting point is that if you ask AI to security audit a codebase it can find the vulnerabilities. But the AI "forget security" when writing new features. This is probably caused by training data and RL.
5 days ago
An interesting point is that if you ask AI to security audit a codebase it can find the vulnerabilities. But the AI "forget security" when writing new features. This is probably caused by training data and RL.
Using LLMs to code is rapidly converging into "programming with natural language" with all drawbacks Dijkstra foresaw and none of the magic big tech promised.