Briefings In-depth technical guides on the latest AI concepts
Liquid Neural Networks
Exploring dynamic neural networks that adapt continuously over time
Spiking Neural Networks
Is this the next frontier in neural network architectures?
Hebbian Learning
Understanding the neural basis of learning and memory.
Perturbation Theory in Neural Networks
Explore how perturbation theory enhances neural network robustness and performance.
This Week’s News
-
Anthropic launches Claude Science, an AI workbench for scientific research that connects Claude to more than 60 scientific databases, supports multi-agent research pipelines, and includes citation and calculation checking for reproducible analysis. [Link]
-
Anthropic releases Claude Sonnet 5 as a cheaper model for agentic work, making it the default for free and Pro users while promising near-Opus performance for coding, tool use, and autonomous workflows. [Link]
-
The US government lifts export controls on Anthropic’s Fable 5 and Mythos 5 models after Anthropic agrees to detect security risks, collaborate on future releases, and alert officials to malicious activity. [Link]
-
The US government is reportedly negotiating voluntary standards for advanced AI model releases with OpenAI, Anthropic, Google, and other companies, including benchmarks, timelines, and access rules. [Link]
-
OpenAI and Broadcom unveil Jalapeño, OpenAI’s first custom LLM inference chip, designed from scratch for AI serving workloads and reportedly developed from initial design to tape-out in nine months. [Link]
-
Google brings Gemini Spark, its agentic assistant, to macOS, adding local file access, Google Tasks and Keep support, third-party app integrations, real-time tracking, and custom MCP connections. [Link]
-
Meta is reportedly building Meta Compute to sell excess AI infrastructure, turning its huge data-center investment into a cloud business for AI compute and hosted model access. [Link]
-
Google introduces Nano Banana 2 Lite for faster, cheaper image generation, producing images in about four seconds at $0.034 per 1,000 images for high-volume creative workflows. [Link]
-
California and Anthropic strike a deal giving state agencies discounted Claude access, including training and support for drafting documents, analysing information, and improving public services. [Link]
-
New consumer spending data suggests Claude is gaining ground among paying AI users, with US consumer revenue up about 75% since January and Claude course demand rising sharply among self-directed learners. [Link]
-
The EU adopts AI Act amendments delaying high-risk compliance deadlines, broadening SME-style exemptions and adding new prohibitions on AI-generated CSAM and non-consensual intimate imagery. [Link]
-
GitHub adds Kimi K2.7 Code as Copilot’s first open-weight model option, giving developers a lower-cost coding model in the Copilot picker under usage-based billing. [Link]
Podcasts Your two favorite Deep Dive hosts discuss AI in depth, courtesy of NotebookLM
Primers Introductory texts on foundational AI concepts
Getting Started with Azure and OpenAI
How to use OpenAI's language models in Azure.
Neurosymbolic AI
Combining neural networks with symbolic reasoning.
Symbolic Artificial Intelligence
A comprehensive introduction to Symbolic AI.
Unleashing the Power of RAG with AWS Bedrock
An easy-to-understand guide on how to create a RAG application using AWS Bedrock.
Propositional and Predicate Logic: A Primer
A comprehensive primer on the theory and application of formal logic.
