Matthew Gunton

Matthew Gunton @matthewjgunton

LLMs are autoregressive models You type in a prompt and they give you back output tokens one at a time What differentiates one run of the LLM from the other are the attention intermediary results (and outputs) that have been calculated to your exact prompt and the output tokens https://t.co/0zjWwn3jk1

Aug 2, 2026 · ♥ 2

Matthew Gunton @matthewjgunton

Why is this graph so important? Interactivity curves show you a fundamental tradeoff that happens when you serve AI Tokens The more requests that you handle at once (your throughput), the slower each request appears to the user (latency) This graph shows the tradeoff that https://t.co/FT8p9ruNqs

Jul 31, 2026 · ♥ 11 · ⇄ 1

Matthew Gunton @matthewjgunton

AI needs a lot of electricity As we add more of it to the grid (and behind the meter) it's worth understanding some basics around how we measure capacity on the grid I want to go over ELCC (Effective Load Carrying Capacity) You see the grid needs to be very careful with https://t.co/EP1WwfqSLs

Jul 30, 2026 · ♥ 4