EPISODE · Apr 18, 2026 · 16 MIN
Inside the Black Box
from Prompted: Tech Talks
Large language models went from an academic paper to global infrastructure in under three years. This episode covers what they actually are, how they work under the hood, and the key concepts — context windows, tool-use, reasoning models — that shape what these systems can and can't do. Part one of two; part two covers the model comparisons. IN THIS EPISODE • Why LLMs became infrastructure so fast — the numbers behind the adoption curve • The timeline: from the 2017 transformer paper to 300 million weekly users • How LLMs actually work: tokens, parameters, and why it's a probability engine, not a search engine • Context windows, tool-use protocols, custom models, and reasoning models explained • The capability debate: Karpathy and Sutskever vs. Yann LeCun — and why neither side has won • The open questions: training data supply, energy footprint, and accountability as AI goes agentic SOURCES • Vaswani et al. — "Attention Is All You Need," Google Brain (2017) — https://arxiv.org/abs/1706.03762 • OpenAI — 300M weekly active users (December 2024) — https://www.cnbc.com/2024/12/04/openais-active-user-count-soars-to-300-million-people-per-week.html • GitHub/Microsoft — Copilot 4.7M paid subscribers, up 75% YoY — Microsoft FY26 Q2 Earnings (January 2026) — https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q2 • DeepSeek — R1 model release (January 2025) — https://github.com/deepseek-ai/DeepSeek-R1 • Jensen Huang — World Economic Forum, Davos (January 2026) — https://www.weforum.org/stories/2026/01/nvidia-ceo-jensen-huang-on-the-future-of-ai/ • Jensen Huang — CES 2026 keynote — https://blogs.nvidia.com/blog/2026-ces-special-presentation/ • Andrej Karpathy — "Intro to Large Language Models" lecture (2023) — https://www.youtube.com/watch?v=zjkBMFhNj_g • Ilya Sutskever — NeurIPS 2024 keynote: "Sequence to Sequence Learning with Neural Networks: What a Decade" — https://www.youtube.com/watch?v=1yvBqasHLZs • Dario Amodei — "Machines of Loving Grace" (October 2024) — https://darioamodei.com/machines-of-loving-grace • Yann LeCun — TIME magazine interview (January 2024) — https://time.com/6694432/yann-lecun-meta-ai-interview/ • Yann LeCun — Lex Fridman Podcast #416 (March 2024) — https://lexfridman.com/yann-lecun-3/ • Anthropic — Model Context Protocol (MCP), open standard (November 2024) — https://www.anthropic.com/news/model-context-protocol
What this episode covers
Large language models went from an academic paper to global infrastructure in under three years. This episode covers what they actually are, how they work under the hood, and the key concepts — context windows, tool-use, reasoning models — that shape what these systems can and can't do. Part one of two; part two covers the model comparisons. IN THIS EPISODE • Why LLMs became infrastructure so fast — the numbers behind the adoption curve • The timeline: from the 2017 transformer paper to 300 million weekly users • How LLMs actually work: tokens, parameters, and why it's a probability engine, not a search engine • Context windows, tool-use protocols, custom models, and reasoning models explained • The capability debate: Karpathy and Sutskever vs. Yann LeCun — and why neither side has won • The open questions: training data supply, energy footprint, and accountability as AI goes agentic SOURCES • Vaswani et al. — "Attention Is All You Need," Google Brain (2017) — https://arxiv.org/abs/1706.03762 • OpenAI — 300M weekly active users (December 2024) — https://www.cnbc.com/2024/12/04/openais-active-user-count-soars-to-300-million-people-per-week.html • GitHub/Microsoft — Copilot 4.7M paid subscribers, up 75% YoY — Microsoft FY26 Q2 Earnings (January 2026) — https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q2 • DeepSeek — R1 model release (January 2025) — https://github.com/deepseek-ai/DeepSeek-R1 • Jensen Huang — World Economic Forum, Davos (January 2026) — https://www.weforum.org/stories/2026/01/nvidia-ceo-jensen-huang-on-the-future-of-ai/ • Jensen Huang — CES 2026 keynote — https://blogs.nvidia.com/blog/2026-ces-special-presentation/ • Andrej Karpathy — "Intro to Large Language Models" lecture (2023) — https://www.youtube.com/watch?v=zjkBMFhNj_g • Ilya Sutskever — NeurIPS 2024 keynote: "Sequence to Sequence Learning with Neural Networks: What a Decade" — https://www.youtube.com/watch?v=1yvBqasHLZs • Dario Amodei — "Machines of Loving Grace" (October 2024) — https://darioamodei.com/machines-of-loving-grace • Yann LeCun — TIME magazine interview (January 2024) — https://time.com/6694432/yann-lecun-meta-ai-interview/ • Yann LeCun — Lex Fridman Podcast #416 (March 2024) — https://lexfridman.com/yann-lecun-3/ • Anthropic — Model Context Protocol (MCP), open standard (November 2024) — https://www.anthropic.com/news/model-context-protocol
NOW PLAYING
Inside the Black Box
No transcript for this episode yet
Similar Episodes
Feb 1, 2025 ·168m
Aug 7, 2024 ·58m