Researchers from Microsoft and Georgia Tech have introduced VCoder, a method that enhances Multimodal Large Language Models’ (MLLMs) object perception abilities. By integrating additional perception modalities, VCoder significantly improves model performance on vision-language tasks, particularly in accurately counting and identifying objects within visual scenes. This innovative approach opens new avenues for refining and optimizing MLLMs’…

The New York Times has filed a lawsuit against OpenAI and Microsoft, alleging copyright infringement through their use of NYT articles to train AI models. The lawsuit asserts that AI-generated responses using NYT content deprive the company of revenue and damages its reputation. If successful, the lawsuit could impact the AI industry and journalism. (Summary:…

Learn to incorporate Llama Guard into RAG pipelines for moderating LLM inputs/outputs and combating prompt injection. Find more details on Towards Data Science.

The rise of large language models driven by artificial intelligence has reshaped natural language processing. Post-training quantization (PTQ) presents a challenge in deploying these models, with optimization choices during pre-training significantly impacting quantization performance. Cohere AI’s research delves into these intricacies, challenging the belief that quantization sensitivity is solely determined by model scale. The study’s…

Privacy in machine learning models has become a critical concern due to Membership Inference Attacks (MIA). The new Relative Membership Inference Attack (RMIA) method, developed by researchers at the National University of Singapore, demonstrates its superiority in identifying membership within machine learning models, offering practical and scalable privacy risk analysis. For more information, visit the…

The NVIDIA 2024 GTC AI conference unites industry influencers in AI and accelerated computing. The in-person event, taking place from March 18-21, 2024, at the San Jose Convention Center, will feature workshops, networking opportunities, and presentations from tech leaders. The event promises to showcase the latest NVIDIA technologies, while offering insightful discussions and hands-on workshops.…

President Biden issued an executive order tasking NIST with researching AI model safety. RAND Corporation’s influence on NIST is under scrutiny due to its advisory role in shaping the order. Concerns have been raised about NIST’s outsourcing of AI safety research, particularly related to organizations like RAND, and its potential impact on AI regulation.

Autonomous driving technology combines AI, machine learning, and sensors to create vehicles capable of human-like decision making. DriveLM, a new model, employs Vision-Language Models for autonomous driving, demonstrating superior adaptability in handling complex driving scenarios. This approach represents a significant advancement in enhancing vehicle perception and decision-making, potentially revolutionizing autonomous driving technology.

MIT, MyShell.ai, and Tsinghua University researchers have developed OpenVoice, an open-source instant voice cloning method. It overcomes voice cloning challenges by enabling flexible voice style control and zero-shot cross-lingual cloning. OpenVoice can replicate a voice, generate speech in multiple languages, control voice styles, and accurately clone the reference speaker’s tone color.

Midjourney has released V6 of its AI image-generating model, introducing the ability to add text to images, along with significant detail and realism upgrades. Founder David Holz highlighted the model’s capability to produce more lifelike imagery. V6 requires more explicit prompts, offers longer detailed prompts, and has enhanced image remixing and upscaling. The release has…