Skip to content
DevMeme
4057 of 7590
AI ML Post #4423 · source on Telegram

AI years feel like dog years since GPT-3 landed on arXiv

Description

Single-panel Tintin - style comic: on the left, Tintin sits blushing at a bar; center, a bearded man in a blue sweater (face blurred) leans on the counter; on the right, Snowy the dog pants beside a half-full pint glass. Large speech bubble from the bearded man reads, “My God - what a decade in AI it's been since GPT-3 hit Arxiv !”. A smaller reply bubble from Tintin states, “That was 2 years ago.” Background is a flat olive color, and the panel is framed with a white border; tiny “meme/dev-meme” watermark appears bottom left. The joke riffs on the breakneck pace and hype of modern AI research - GPT-3’s 2020 arXiv paper already feels ancient, highlighting how quickly model capabilities, tooling, and industry expectations evolve for engineers tracking large language models

Comments

6
Anonymous ★ Top Pick AI moves so fast that by the time procurement signs off on the A100s for our GPT-3 pilot, the model’s considered “legacy” and the PyTorch version it needs has slipped into LTS
  1. Anonymous ★ Top Pick

    AI moves so fast that by the time procurement signs off on the A100s for our GPT-3 pilot, the model’s considered “legacy” and the PyTorch version it needs has slipped into LTS

  2. Anonymous

    Remember when we thought microservices adoption was fast? AI research papers now have a half-life shorter than our sprint cycles, and every standup starts with 'So, which foundation model got deprecated while we were sleeping?'

  3. Anonymous

    In AI years, GPT-3's arXiv release is basically the Cambrian explosion - ancient history where we were still figuring out if 175B parameters was 'too many.' Now we're casually discussing trillion-parameter models while GPT-3 sits in production like COBOL: technically legacy, still running half the internet, and nobody wants to admit how recently it was cutting-edge

  4. Anonymous

    GPT-3 dropped two years back, birthing the LLM era - yet our enterprise monoliths from Y2K still demand COBOL priests

  5. Anonymous

    Two years on the calendar, but in engineering time that’s five SDK rewrites, three eval pipelines, and one deprecation notice the day after GA

  6. Anonymous

    Two years since GPT‑3 and it already feels like a decade: four API renames, three vector DBs, and our “temporary prompt” somehow has an SLA

Use J and K for navigation