SSanjanainunder-the-hood-ai.hashnode.dev·Jul 27 · 19 min readFrom Prompt to Token: Inside Modern LLM InferenceWhy This Matters Modern LLMs can remember thousands of previous tokens while continuing a conversation. But if every new response needs to consider everything we have said so far , why doesn't the mo00
AAditibharadwajinaditibharadwaj.hashnode.dev·Jul 23 · 7 min readDo Transformers Understand Order ?Self-attention is incredibly powerful. It can compare every word with every other word in parallel but it has one surprising blind spot, it has no inherent understanding of sequence order. Why Is Atte00
RRishiinrishi2220.hashnode.dev·Jul 22 · 36 min readTowards the Modern Transformer ArchitectureHi y'all, Long time no see! Last post was last year. Everyone has likely read Attention Is All You Need, and understood everything around self-attention, and mapped out how encoders and decoders talk 11K
SPSaurabh Prajapatiinblog.thitainfo.com·Jul 22 · 20 min readHow ChatGPT Understands Your Questions: A Deep Dive into LLMs, Tokenization, and TransformersHow ChatGPT Understands Your Questions: A Deep Dive into LLMs, Tokenization, and Transformers TL;DR: ChatGPT doesn't "read" your question the way you do. It converts text into numbers, processes thos00
AJAman Jainincurious-pm.hashnode.dev·Jul 20 · 7 min readDoes a Wider Embedding Space Help More on Diverse Data?TL;DR Increasing the embedding dimension from 32 to 64 improved validation loss in both the low- and high-diversity training conditions. Most of the improvement appeared between 32 and 48 dimensions00
AJAman Jainincurious-pm.hashnode.dev·Jul 20 · 11 min readDo Bigger GPT Models Always Perform Better?TL;DR: What I tested and observed Experiment 1 — Different sizes, fixed tokens: I increased model width, depth, and total parameter count while giving every model the same 200,192 training tokens. In00
PKPulkit Kushwahainpulkitkushwaha.hashnode.dev·Jul 16 · 10 min readWhy AI Suddenly Exploded in 2022 Even Though We Had the Algorithms Decades AgoWhy AI Suddenly Exploded in 2022 Even Though We Had the Algorithms Decades Ago The technology behind ChatGPT wasn't new. Neural networks existed before the internet. Backpropagation is older than many50
CPChai Planetinthe-tech-inside.hashnode.dev·Jul 14 · 7 min readEvery AI prompt Triggers an Invisible FactoryYou type a question. Hit enter. Get an answer in seconds. But what you don't see is the invisible factory that just ran to generate those words. A massive, power-hungry manufacturing plant that spun u00
KKKrishna Kumar Yadavindevintelligence.hashnode.dev·Jul 13 · 25 min readHow ChatGPT Understands Your Questions?1. What is an LLM? A. What LLM stands for LLM stands for Large Language Model. Let’s break the name into three parts. Large The word Large doesn’t simply mean the model occupies a lot of storage. It m00
CDCanary Digitalincanary-digital.hashnode.dev·Jul 14 · 1 min readThe History of AI Models - Part 3: The Transformer Era & LLMs By 2017, models like RNNs and LSTMs were experiencing serious bottlenecks in understanding long texts and parallel processing. That is, until that historic paper by Google researchers was published...00
