Those wanting a firmer grip on chatbots, then, may have to explore more underhand techniques, such as the one discovered by two computer-science researchers at Harvard University. They’ve demonstrated how chatbots can be tactically controlled by deploying something as simple as a carefully written string of text. This “strategic text sequence” looks like a nonsensical series of characters – all random letters and punctuation – but is actually a delicate command that can strong-arm chatbots into generating a specific response. Not part of a programming language, it’s derived using an algorithm that iteratively develops text sequences that encourage LLMs to ignore their safety guardrails – and steer them towards particular outputs.
| Artificial Intelligence |
| Research Data Curation and Management Works |
| Digital Curation and Digital Preservation Works |
| Open Access Works |
| Digital Scholarship |
