GPT-4 for catching GPT-4’s mistakes:
SAFETY
-
AI Leaderboard Updates, Olympic Vocals, and Security Concerns
By
–
Top stories in AI today: -Hugging Face updates Open LLM Leaderboard
-NBC rolls out AI vocals for Olympic recaps
-Enhance videos with Krea AI upscaling
-Rabbit R1 hit with major security flaw
-5 new AI tools & 4 new AI jobs Read more: http://
therundown.ai/p/ai-gets-lead
erboard-shakeup
… -
AI’s Main Limitation: Learning Only Before Deployment
By
–
The #1 limitation of current AIs is that they only learn before they’re deployed.
-
GenAI Hallucinations: Why Models Fail Without Information
By
–
3. Hallucinations make GenAI applications unusable Models hallucinate because they are probabilistic. However, a model is much more likely to hallucinate when it doesn’t have access to the right information. Multiple studies have shown that hallucinations can be
-
Model Merging Preserves Safety Alignment Better
By
–
Model Merging and Safety Alignment New paper looking into how model merging poorly preserves safety alignment. The authors modify EvoMM and LM-Cocktail to balance performance on safety data and domain-specific data. They show that this safety-aware merging approach can
-
ML Security: Why Attacking Systems Validates Their Safety
By
–
*Must read* for anyone interested in ML security, by Nicholas Carlini. Attacks are the only way we know whether or not a purportedly secure system actually is. Moreover, I consider personal attacks like this unacceptable in my research communities. https://
nicholas.carlini.com/writing/2024/w
hy-i-attack.html
… -
FBI AGI: Real-Time Social Media Predator Detection Tool
By
–
In 1st: FBI AGI A real-time incrimination tool against predators on social media. Built by Alex Sima & Apurva Mishra
-
Hasty AI Deployment Strategy Backfires Against Google Competition
By
–
Por un lado, se entiende la dificultad de desplegar de forma segura una tecnología tan inexplorada como esta. Por el otro, qué ridículo es querer chulear con prisas para opacar el evento de Google si luego no tienes los deberes hechos. Se deben escuchar risas desde Google.
-
OpenAI Delays Voice Mode Launch by One Month for Safety
By
–
Pues nada… El modo voz se retrasa un mes. OpenAI quería empezar la alpha para esta fecha pero necesitan más tiempo para asegurar que el sistema sea seguro y escalable a millones de usuarios. El mes de retraso mas lo que será una alpha progresiva: nos vamos hasta otoño