Section C
Papers explained, benchmarks, datasets and the science behind AI.
The 2015 ResNet paper solved the degradation problem with residual connections and won ImageNet. How skip connections work and why every transformer uses them.
AI Desk3 min read
In 2020, Google showed that a plain transformer applied to image patches can rival convolutional networks. How ViT works, why it needs lots of data and what it enabled.
AI Desk4 min read
CLIP learned from 400 million image-caption pairs to match pictures with text, enabling zero-shot image classification and powering text-to-image generators.
AI Desk4 min read
The 2023 LLaVA paper connected a vision encoder to an open language model and trained it on AI-generated visual instructions, creating an open multimodal assistant.
AI Desk3 min read
DeepMind's deep Q-network learned dozens of Atari games from raw screen pixels and scores alone. How Q-learning, experience replay and target networks made it work.
AI Desk3 min read
Proximal Policy Optimization made policy-gradient training simpler and more stable by limiting how far each update can move. How it works and why it is everywhere.
AI Desk3 min read
Anthropic used sparse autoencoders to decompose a model's neurons into thousands of interpretable features. What superposition is, what they found and why it matters.
AI Desk3 min read
Microsoft researchers argued that GPT-4 showed early signs of general intelligence. What they tested, why critics objected and what it teaches about AGI claims.
AI Desk3 min read
Google DeepMind proposed rating AI systems by performance and generality, like levels of self-driving cars. The six levels, the principles behind them and the debate.
AI Desk3 min read
EleutherAI's 825 GB Pile combined 22 diverse text sources into an open training set. What it contained, why diversity mattered and the copyright questions it raised.
AI Desk3 min read
Most language models learn from filtered web crawls. How Common Crawl works, how FineWeb cleaned 15 trillion tokens, and why filtering choices shape model quality.
AI Desk3 min read
The most important tool in AI engineering: how to collect test cases, define success criteria, score outputs automatically and by hand, and run evals on every change.
AI Desk3 min read
Language models can score other models' answers at scale. How LLM-as-a-judge works, its known biases, and how to design rubrics and checks that make it reliable.
AI Desk3 min read
Robots that follow instructions like "pick up the extinct animal": how vision-language-action models such as RT-2 connect web knowledge to physical actions.
AI Desk3 min read
The sensors behind autonomous vehicles, what each does best, how AI fuses them into a picture of the road, and the debate between camera-only and multi-sensor designs.
AI Desk3 min read
Academic integrity, AI disclosure rules, research use, teaching AI literacy and institutional AI tools: how higher education is adapting to ChatGPT-era students.
AI Desk3 min read
Weather forecasting, grid optimization, materials discovery, emissions monitoring and climate adaptation: where AI makes a measurable difference for the environment.
AI Desk3 min read
Most AI is built for English and a few major languages. Why low-resource languages lag, what that means for billions of people, and the efforts to close the gap.
AI Desk3 min read