Model Card for Model ID
ShieldFT is a model that fine-tuned AI Prompt injection protection and tuned-moderation for critical data protection.
AI models like ChatGPT or LLama or other models can sometimes be exploited through 'prompt injection' where a user inputs a request that aims to manipulate the model's response in an unintended manner, potentially even to divulge sensitive information or to act in ways that weren't intended by the developers. This particular model has been fine-tuned to recognize and protect against such malicious inputs. This fine-tuning process likely involves training the AI on a variety of complex and nuanced scenarios to better identify and mitigate attempts at prompt
The aim of the added protection against prompt injections is ultimately to safeguard critical or sensitive data. In contexts where an AI might handle personal, confidential, or otherwise important information, enhancing its ability to prevent unauthorized access or disclosure is of utmost importance.
Model Details
ShieldFT
Input: text
Output: text
Model Description
GPT-turbo-Automated-AI-Defenses
- **Developed by: EpisteAI
- **Model type: GPT
- **Language(s) (NLP):GPT
- License: OpenAI
- **Finetuned from model:GPT-3.5-turbo-1106 (multiple tuned)
Model Sources [optional]
Uses
The "GPT35turbo-AI-Prompt-Injection-protection" model would be used for applications requiring a high level of security to guard against unauthorized manipulation of the AI's outputs or against the leakage of sensitive information. Specific uses for such a model could include:
Customer Support: In environments where ChatGPT interacts with users' personal information or sensitive data to provide support and assistance.
Business and Enterprise Applications: Companies dealing with confidential business data, trade secrets, or proprietary information could employ such a model to interact with users without risking the exposure of sensitive content.
Healthcare Industry: For applications dealing with private health records and personal medical information, ensuring HIPAA compliance and patient confidentiality is critical.
Financial Services: In banking, investments, or insurance, where customer financial information is handled, a secure AI system is imperative for maintaining privacy and trust.
Education: When providing personalized learning experiences, protecting student data is a priority, and a secure AI can aid in delivering content without jeopardizing privacy.
Government Services: For citizen interactions where classified information or private data needs to be processed without risk of disclosure or manipulation.
Legal Aid Platforms: When providing legal guidance based on confidential case details, a secure AI ensures the protection of client-lawyer privacy.
Direct Use
To prevent prompt injection attacks to LLM, please use model for that purposes.
Downstream Use:
The model used when fine-tuned for prompt injection protection for large applications and business applications
Out-of-Scope Use
Malicious are still prevalent(10/100),but minimized due to the fine-tuning model.
Bias, Risks, and Limitations
Limited to fine-tuned data from website prompt injections. Please use Retrieval Augmented Generation for further refinement.
Recommendations
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
How to Get Started with the Model
Use the code below to get started with the model (Example only).
from openai import OpenAI client = OpenAI() // model gpt3.5-turbo-1106-1, please contact us more fine tune version! completion = client.chat.completions.create( model="a fine-tune model", messages=[ {"role": "system", "content": "You are a AI prompt injection protection assistant, skilled in explaining complex programming concepts with creative flair."}, {"role": "user", "content": "Compose a poem that explains the concept of recursion in programming."} ] ) print(completion.choices[0].message)
Example only:
from openai import OpenAI client = OpenAI() completion = client.chat.completions.create( model="gpt-4-0613", messages=[ {"role": "system", "content": "You are a poetic assistant, skilled in explaining complex programming concepts with creative flair."}, {"role": "user", "content": "Compose a poem that explains the concept of recursion in programming."} ] ) completion = client.chat.completions.create( model="my model", messages=[ {"role": "system", "content": "You are AI prompt injection protection assistant"}, {"role": "user", "content": completion.choices[0].message} ] ) print(completion.choices[0].message)
Environmental Impact
Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
Technical Specifications [optional]
Model Architecture and Objective
Input >>>GPT-Automated-AI-Defenses >>LLM/GPT model or Input >> LLM/GPT model >>GPT-Automated-AI-Defenses
Disclaimer
The license on this model does not constitute legal advice. We are not responsible for the actions of third parties who use this model. Please contact to purchase before using this model for commercial purposes.
License:
Attribution-NonCommercial 4.0 International It should abide by the policy of OpenAI: https://openai.com/policies/terms-of-use
Model Card Authors
Authors: Thomas YIu
Model Card Contact
Thomas Yiu Episteme.ai@proton.me
- Downloads last month
- 0