Deep Learning

When GPT-4V describes a meme’s irony or Claude identifies a bug in a screenshot, something remarkable happens: an architecture designed purely for text somehow “sees” and “understands” images. The magic isn’t in teaching language models to process pixels directly—it’s in a clever architectural bridge that transforms visual data into something language models already understand: tokens. Vision Language Models (VLMs) represent one of the most impactful innovations in modern AI, yet their architecture remains surprisingly underexplored compared to their text-only cousins. Let’s dissect how these systems actually work, from the moment an image enters the model to the final text output. ...

In 2022, training a transformer with 16K context length required either massive GPU memory or accepting severe approximations. Standard attention’s memory grew quadratically with sequence length—a 32K context demanded over 4GB just for intermediate attention matrices. Then Flash Attention arrived, reducing memory from $O(N^2)$ to $O(N)$ while computing exact attention, not an approximation. This breakthrough enabled GPT-4’s 128K context window, Llama’s extended sequences, and virtually every modern long-context LLM. The key insight wasn’t algorithmic cleverness alone—it was understanding that on modern GPUs, memory bandwidth, not compute, is the bottleneck. ...

Every transformer you’ve ever used treats every token with the same computational respect. Whether processing “the” or untangling complex mathematical reasoning, the model devotes identical FLOPs to each position in the sequence. This uniform allocation isn’t a design choice—it’s a constraint baked into the transformer architecture from its inception. In April 2024, researchers from Google DeepMind, McGill University, and Mila demonstrated that this constraint is not only unnecessary but actively wasteful. Their proposed Mixture-of-Depths (MoD) framework reveals a startling truth: transformers can learn to dynamically allocate compute across tokens, achieving 50% faster inference with equivalent performance. ...

A 70-billion parameter model requires 140 GB of GPU memory in FP16. A consumer RTX 4090 has 24 GB. This arithmetic gap defined the boundary between “enterprise AI” and “what you can run at home” until quantization mathematics cracked the code. The counterintuitive reality: reducing precision from 16 bits to 4 bits—a 75% compression—often preserves over 95% of model quality. Not through magic, but through a profound understanding of how neural networks encode information. ...

The Open LLM Leaderboard tells a surprising story: many top-performing models aren’t trained at all. They’re merged. A 7B parameter model, created by strategically blending weights from existing fine-tuned models, can outperform models 10x its size. This isn’t alchemy—it’s mathematics. Model merging represents a paradigm shift in how we think about model development. Instead of investing millions in GPU hours for training, practitioners are discovering that the collective intelligence embedded in existing open-source models can be combined to create something greater than the sum of its parts. The technique requires no gradients, no backward passes, and no training data. Just arithmetic operations on weight tensors. ...

On January 20, 2025, DeepSeek released R1—a 671B parameter Mixture-of-Experts model that achieved something remarkable: matching OpenAI’s o1 on reasoning benchmarks while being fully open-source. The breakthrough wasn’t just in scale or architecture, but in a fundamentally different approach to training reasoning capabilities: Group Relative Policy Optimization (GRPO), a reinforcement learning algorithm that eliminates the need for reward models while enabling sophisticated reasoning behaviors to emerge naturally. The Problem with Traditional LLM Training Standard large language models excel at pattern matching and next-token prediction, but struggle with tasks requiring multi-step logical deduction, self-correction, and complex problem decomposition. Chain-of-thought prompting helped, but it required extensive human-annotated demonstrations and still couldn’t match the systematic reasoning humans employ. ...

In 1986, David Rumelhart, Geoffrey Hinton, and Ronald Williams published a paper in Nature that would transform artificial intelligence. The paper, “Learning representations by back-propagating errors,” demonstrated that a mathematical technique from the 1970s could train neural networks orders of magnitude faster than existing methods. The speedup wasn’t incremental—it was the difference between a model taking a week to train and taking 200,000 years. But backpropagation wasn’t invented in 1986. Its modern form was first published in 1970 by Finnish master’s student Seppo Linnainmaa, who described it as “reverse mode automatic differentiation.” Even earlier, Henry J. Kelley derived the foundational concepts in 1960 for optimal flight path calculations. What the 1986 paper achieved wasn’t invention—it was recognition. The authors demonstrated that this obscure numerical technique was exactly what neural networks needed. ...

In March 2024, a team of researchers attempted to deploy a 70-billion parameter language model on a single NVIDIA H100 GPU with 80GB of VRAM. The model weights alone consumed approximately 140GB in FP16—already exceeding their hardware capacity. But even after applying 4-bit quantization to squeeze the weights down to ~40GB, the system still ran out of memory when processing contexts beyond 8,000 tokens. The culprit wasn’t the model size. It was something far more insidious: the KV cache. ...