Putting up numbers

Claude 3.7 Sonnet seems great, so it's still surprising the model can struggle with our Reverse QA task: generating accurate questions for input answers (which doesn't require that much reasoning imo)

LLMs struggle to generalize when counting r's in strawberry but have no problem generalizing when it comes to bullying 😭

What happens if we fine-tune an LLM on more PII? We find that PII that wasn’t previously extracted gets extracted after fine-tuning on *other* PII. This could have implications for earlier seen data (e.g. during post-training or further fine-tuning). 🧵 arxiv.org/pdf/2502.15680

we are pr- uh... we are happy to announce our brand refresh, reflecting our role in dimensionality reduction and translation of complex AI/ML topics to an AI engineering audience

Today we are announcing ARC-AGI-2, an unsaturated frontier AGI benchmark that challenges AI reasoning systems (same relative ease for humans). Grand Prize: 85%, ~$0.42/task efficiency Current Performance: * Base LLMs: 0% * Reasoning Systems: <4%

i have a story for you all:

Hiring a Marketer for Learn Prompting! We are looking for a scrappy marketer to join our team. We have grown tremendously quickly with a small team and have served 3,000,000+ users in our mission to teach the world prompt engineering...

Our new prompt hacking resources collection is live! We've compiled a big list for anyone interested in jailbreaking, AI red teaming, or prompt injection. We covered: - Blogs - Communities - Courses - Events - Jailbreaks - YouTube videos Please star, contribute, and share!

"The AIRT Masterclass was one of the best, most practical hands-on trainings I've taken in a long time." Just wrapped up our 2nd cohort of our AI Red Teaming Masterclass, and we're just getting started! If you're looking to master AI Security, we'd like to see you in our next

We are hiring a technical writer to help write documentation on prompt engineering techniques. Please reach out if this is you: - 3+ years of technical AI writing - Research background - Passion for education

🌱 We are growing! We are currently hiring for a GTM lead to help lead enterprise sales for our AI Red Teaming Services. We’re quickly becoming the go-to provider for a variety of AI Red Teaming services, and the industry-standard training provider for careers within AI

This week, I learned that prompt injection was invented earlier than I previously thought👀 Learn Prompting #AIRTP+

We are looking for a cracked NextJS+Supabase dev to join our team. We are building out HackAPrompt, the largest AI Red-Teaming competition platform ever. Read more below👇

Woooo

Looks super cool!

Opus 4 just delivered the best AI web dev performance I've ever seen! 🤩 I challenged Opus to comprehensively research yours truly and then design and build a personal website to aggregate the most interesting tidbits. Not only did Opus one-shot a MASSIVE html monolith that imo