the-decoder6 天
AutoDAN-Turbo autonomously develops jailbreak strategies to bypass language model safeguards
A team of researchers from US universities and Nvidia has created AutoDAN-Turbo, a system that automatically finds ways to bypass safeguards in large language models. AutoDAN-Turbo works by ...
the-decoder3 天
German education leaders calls for systematic integration of AI in the classroom
Germany has adopted new guidelines for the use of artificial intelligence in education. Christine Streichert-Clivot, who leads the national body that coordinates education policy across Germany (the ...
the-decoder5 天
AI model simulates Counter-Strike with 10 FPS on a single RTX 3090
Researchers have created an AI model that can simulate a janky version of Counter-Strike: Global Offensive (CS:GO) inside a neural network. The model, called DIAMOND (Diffusion for World Modelling), ...
the-decoder4 天
REPA accelerates diffusion model training by a factor of 17.5
A new technique called REPA dramatically accelerates the training of AI image generation models. The method leverages insights from self-supervised image models to improve both speed and quality. REPA ...
the-decoder3 天
Mistral AI launches Ministral 8B, a much more capable successor to its popular Mistral 7B LLM
French AI startup Mistral AI has announced two new language models designed for use on edge devices and in edge computing scenarios. The new models, called Ministral 3B and Ministral 8B, are part of ...
the-decoder6 天
'OCR 2.0' model converts images of text, formulas, notes, and shapes into editable text
Researchers have created a new universal optical character recognition (OCR) model called GOT (General OCR Theory). Their paper introduces the concept of OCR 2.0, which aims to combine the strengths ...
the-decoder3 天
AI software fixes eye contact in videos for 10 cents a minute
Sieve, an AI startup, has launched a new API that claims to fix eye contact in videos with a single call. The company says its technology can adjust gaze direction automatically in real-time to make ...
the-decoder4 天
Meta researchers develop method to make AI models "think" before answering
Scientists from Meta, UC Berkeley, and NYU have created a new technique to improve how large language models (LLMs) approach general tasks. Called "Thought Preference Optimization" (TPO), the method ...
the-decoder4 天
If you still trust online video, take a look at TANGO
A new AI system called TANGO can generate realistic videos of people gesturing and moving to match any audio recording. This technology could make it even harder to spot fake videos online. TANGO ...
the-decoder4 天
New York Times takes legal action against LLM search engine Perplexity
The New York Times (NYT) is taking legal action against AI startup Perplexity. The publisher accuses the company of using its content without permission for AI-powered search results. According to a ...
the-decoder4 天
OpenAI says ChatGPT has much less gender bias than all of us
OpenAI researchers have found that the usernames people choose when interacting with ChatGPT can subtly influence the AI's responses. But overall, this influence is very small and limited to older or ...
the-decoder5 天
ComfyGen AI automates multi-stage text-to-image workflows from simple prompts
Researchers from Nvidia and Tel Aviv University have developed ComfyGen, a new AI method that automates the creation of text-to-image workflows from simple prompts. The system independently selects ...