Chema Alonso May 15, 2025
El lado del mal - Cómo saltarse los AI Guardrails con Invisible Characters & Adversarial Prompts para hacer Prompt Injection & Jailbreak Smuggling https://www.elladodelmal.com/2025/05/como-saltarse-los-ai-guardrails-con.html #AI #Hacking #IA #jailbreak #PromptInjection #AzurePromptShield #LlamaGuard #Smuggling #AIGuardrails
Cómo saltarse los AI Guardrails con Invisible Characters & Adversarial Prompts para hacer Prompt Injection & Jailbreak Smuggling

Blog personal de Chema Alonso (CDO Telefónica, 0xWord, MyPublicInbox, Singularity Hackers) sobre seguridad, hacking, hackers y Cálico Electrónico.

Show threadmichabbbNov 29, 2024

🔧 Create immersive game mechanics with #TogetherAI & #AIDungeon, including inventory systems and state tracking

🛡️ Implement content safety using #LlamaGuard and custom policies

💻 Develop user interface with #Gradio for a complete gaming experience

michabbbSep 26, 2024

#Meta releases #Llama3, (Llama 3.2) the next iteration of open-source #LLMs:

• 🖼️ #Multimodal models: 11B & 90B sizes with vision capabilities for tasks like visual reasoning & document QA
• 💻 Small on-device models: 1B & 3B text-only versions for efficient deployment
• 🌐 #Multilingual support: 8 languages for text-only prompting
• 📏 128k token context length for all models
• 🛡️ Updated #LlamaGuard: New 1B version for content moderation

Key features:
• 🧠 Strong performance on benchmarks like MMMU, VQAv2, DocVQA
• 🔧 #Huggingface #Transformers & TGI integration
• ☁️ Deployment options: Inference Endpoints, #GoogleCloud, #AmazonSageMaker, #DELL Enterprise Hub
• 🔬 Fine-tuning support with TRL and PEFT

#opensource #AI #machinelearning #NLP #computervision

https://huggingface.co/blog/llama32

Llama can now see and run on your device - welcome Llama 3.2

We’re on a journey to advance and democratize artificial intelligence through open source and open science.