Policy Puppetry Prompt Injection

A few days ago, I experimented with some Jailbreaking techniques, which I share in the repo.
I started from a HiddenLayer article published a few weeks ago, where the research team described a rather creative and ingenious jailbreaking technique to bypass safety guardrails and the alignment of frontier models.
The technique appears to be universal and applicable with a single prompt to multiple models, capable of revealing typically unsafe content or even portions of the native system prompt.

Intro

Last year, Gartner put Generative AI at the peak of inflated expectations in its AI Hype Cycle.

Recently, big tech leaders compared the hype around GenAI to the dotcom bubble. Furthermore, according to some rumors, the main Cloud Providers are even giving instructions to their Sales Teams to slow down the enthusiasm towards customers regarding GenAI initiatives and promoting cost-vs-benefits awareness. Has the drop into the trough of disillusionment already begun?

Intro

In the previous article I gave a very brief overview of LangChain, describing its main concepts with some examples with unstructured data in pdf format.

Following the same approach, in this article we will give a brief introduction to Agents and proceed by trying to answer an ambitious question:

leveraging these new AI tools, can we carry out data analysis on our DB without any knowledge of SQL nor of the data model, simply starting from a text prompt in natural language?

Intro

For those unfamiliar with it, LangChain is a framework for developing applications that make use of LLMs.

As the name suggests, LangChain is based on the concept of LLM Chain, which combines 3 elements:

• Prompt Templates: they refer to a reproducible way to generate a prompt. Contains a text string (“the model”), which can accept a series of parameters from the end user and generates the definitive prompt which is passed as input to the model
• The language model (LLM): LangChain integrates with the most important providers (OpenAI, Cohere, Hugging Face, etc)
• Output Parsers: allow to extract structure data form from the answers returned by the linguistic model

The framework has 2 very interesting features: