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

PydanticAI: Revolutionizing Agentic App Development

PydanticAI introduces an evaluation-driven approach to developing agentic applications, addressing challenges like non-determinism and LLM limitations. The framework allows for mock dependencies, enabling developers to build evaluation-driven applications efficiently.

Unlocking the Power of Layer Enhanced Classification (LEC)

New approach LEC effectively classifies content safety violations and prompt injection attacks using hidden states from intermediate Transformer layers. LEC outperforms special-purpose models and GPT-4o, offering a lightweight and efficient solution for businesses to protect against model manipulation.

DIY Dataset: LLM Training 101

Large language models (LLMs) require well-curated datasets for optimal performance. Data preprocessing involves extracting text from diverse sources and filtering for quality, using tools like OCR and regex filters.

Master AdaBoost Regression in C#

AdaBoost.R2 modifies AdaBoost for regression, creating a sequence of decision trees for better predictions. Weighted median enhances accuracy by emphasizing high-confidence tree predictions.

AI: Your Research Hypothesis Helper

MIT researchers developed SciAgents, an AI framework for generating evidence-driven research hypotheses in biologically inspired materials, leveraging graph reasoning methods. Co-authored by Alireza Ghafarollahi and Markus Buehler, the study aims to simulate the collective intelligence of scientists to accelerate the discovery process.

Drones revolutionize warehouse inventory tracking

Corvus Robotics uses autonomous drones for efficient warehouse inventory management, increasing operational speed and accuracy. Co-founder Mohammed Kabir developed the drone platform to navigate warehouses without GPS, revolutionizing inventory tracking.

AI Decodes Whisky Aromas

Artificial intelligence outperforms experts in identifying whisky notes. AI predicts aromas and origin accurately, advancing automated whisky aroma analysis.

MIT Welcomes Frida Polli: Visiting Innovation Scholar

Frida Polli, MIT's new visiting innovation scholar, transitioned from neuroscience to entrepreneurship, co-founding successful AI company pymetrics. Polli's work led to algorithmic bias law, collaboration with Sendhil Mullainathan to bridge behavioral science and computer science at MIT.

UK Technology Access for Blacklisted Chinese AI Chip Firms

Imagination Technologies licenses UK tech to Chinese firms for AI chips in advanced weapons systems, raising national security concerns. Chinese companies Moore Threads and Biren Technology under US export restrictions for developing AI chips for military purposes.

Revolutionizing LLM Safety: NeurIPS 2024 Challenge Insights

NeurIPS 2024 Challenge 2nd prize winner presents a unique approach to effective LLM unlearning without retaining dataset, using reinforcement learning and classifier-free guidance. The competition focused on forcing LLM to generate personal data and protecting it; solution involved supervised finetuning, reinforcement learning, and CFG.

Mastering Soft Actor-Critic RL

Soft Actor-Critic (SAC) is a new off-policy deep RL algorithm addressing stability issues in high-dimensional environments. SAC promotes robustness and exploration in bioengineering systems like de novo drug design.

MIT engineers reach new heights with 3D chip growth

Chip manufacturers are exploring multilayered chip designs to increase computing power. MIT engineers have developed a method to stack high-quality semiconducting layers without bulky silicon substrates, potentially revolutionizing AI hardware.