NEWS IN BRIEF: AI/ML FRESH UPDATES

Get your daily dose of global tech news and stay ahead in the industry! Read more about AI trends and breakthroughs from around the world

Introducing Pixtral 12B on Amazon SageMaker JumpStart

Pixtral 12B, Mistral AI's cutting-edge vision language model, excels in text-only and multimodal tasks, outperforming other models. It features a novel architecture with a 400-million-parameter vision encoder and a 12-billion-parameter transformer decoder, offering high performance and speed for understanding images and documents.

Unveiling Matrix Singular Values

Singular values of matrices can be computed using the SVD technique, but a paper by C. R. Johnson offers a lower bound method for estimating the smallest singular value. Early hybrid aircraft designs combining piston engines with jets were quickly abandoned in favor of pure jet engines due to rapid technological advancements.

Mastering Predicted Probability: Visual Guide & Code Examples

Classification models provide not only answers but also confidence levels through probability scores. Explore how seven basic classifiers calculate and express their prediction certainty visually. Understanding predicted probability is key to interpreting how models make choices with varying levels of confidence.

Unbiased AI: Accuracy Preserved

MIT researchers developed a new technique to improve machine-learning model accuracy for underrepresented groups by removing specific data points. This method addresses hidden biases in training datasets, ensuring fair predictions for all individuals.

Enhancing Presentations with Multimodal Foundation Models

Two approaches to gain insights on multimodal data: embed first, infer later with Amazon Titan Multimodal Embeddings, and infer first, embed later with Anthropic’s Claude 3 Sonnet. Evaluation using SlideVQA dataset, providing concise responses to user questions.

Daniela Rus: John Scott Award Winner

MIT's Daniela Rus receives 2024 John Scott Award for groundbreaking robotics research, redefining the capabilities of robots beyond traditional norms. Rus's work focuses on developing explainable algorithms to create collaborative robots that can solve real-world challenges, emphasizing the synergy between the body and brain for intelligent machines.

McCartney: AI Threatens to Take Over UK Copyright Laws

Paul McCartney warns AI could threaten income streams for creators, calls for laws against mass copyright theft by AI companies. The former Beatle expresses concern over young composers and writers unable to protect their intellectual property from algorithmic models.

Revolutionizing AI Content with Citation Tool

MIT CSAIL researchers developed ContextCite, a tool to enhance trust in AI-generated content by identifying external context sources. This tool helps users verify statements, trace errors back to sources, and detect hallucinated information.

Decoding AI: Making Predictions Understandable

MIT researchers developed a system using large language models to convert complex AI explanations into plain language, improving user understanding. The system evaluates the quality of the narrative, allowing users to trust machine-learning predictions and customize explanations to meet specific needs.

Maximizing LLM Performance: The Power of Hybrid Models

Implementing Speculative and Contrastive Decoding enhances text generation quality and efficiency using large and small language models. Contrastive Decoding prioritizes tokens with the largest probability difference between models for high-quality outputs.

AI Saves Lives: Reducing Stillbirths in Malawi

The only hospital in the country using foetal safety software reduced baby fatalities by 82% in 3 years. A routine ultrasound at Area 25 health centre in Malawi's capital saved Ellen Kaphamtengo's baby from potential asphyxia.