Topics
A closed vocabulary, deliberately small. Each shelf pulls together everything on that subject — the two-minute news item and the twenty-minute teardown sit side by side.
LLMs
Language-model architecture, behaviour and capability.
44 postsLatest
Reasoning
Chain of thought, test-time compute, reasoning models and their limits.
16 postsLatest
Agents
Tool use, agent frameworks, MCP, multi-step autonomy.
32 postsLatest
Genie 4: when Gemini directs the world model, robots get a playground
Interpretability
Features, circuits, probing, activation steering, representation engineering.
2 postsLatest
SMILESGNN: Fusing SMILES and Graphs for Better Toxicity Prediction
Deep learning
Architectures, optimisation, scaling laws, training dynamics.
16 postsLatest
Reinforcement learning
RLHF, RLVR, reward models, policy optimisation.
3 postsLatest
NVIDIA Releases Metatron v1 for Interactive Humanoid Control
Multimodal
Vision, audio, video and cross-modal models.
14 postsLatest
Genie 4: when Gemini directs the world model, robots get a playground
Training
Pre-training, fine-tuning, data, tokenisation, curricula.
10 postsLatest
Inference
Serving, quantisation, speculative decoding, caching, local and on-device.
18 postsLatest
NVIDIA Releases Metatron v1 for Interactive Humanoid Control
Evaluation
Benchmark design, contamination, harnesses, construct validity.
13 postsLatest
AI security
Prompt injection, jailbreaks, agent hijacking, exfiltration, model supply chain.
22 postsLatest
AI safety
Alignment, guardrails, deception, misuse, model behaviour under pressure.
9 postsLatest
Generative AI
Diffusion, image, audio and video generation, and their systems.
6 postsLatest
Genie 4: when Gemini directs the world model, robots get a playground