February 23, 2026
From Transformers to tantrums
Show HN: AI Timeline – 171 LLMs from Transformer (2017) to GPT-5.3 (2026)
A giant AI family tree drops — commenters nitpick, demand light mode, and accuse Codex of “bug cosplay”
TLDR: A massive timeline charts 171 AI models from the 2017 Transformer to a speculative 2026 GPT‑5.3. Commenters nitpicked omissions, demanded light mode and visual charts, and debated whether code‑writing AI makes bugs on purpose—turning a simple history lesson into a lively, useful brawl about usability and trust.
Hacker News drops a mega “AI family album” listing 171 Large Language Models (LLMs — AI that reads and writes text) from the 2017 Transformer to a teased 2026 GPT‑5.3, and the comments instantly took over. First in: NitpickLawyer claiming big omissions like GPT‑J and GPT‑NeoX — the scrappy open models that ran on home hardware — sparking an open‑source pride parade. Maro wanted a visual feast: color‑coded, to‑scale network diagrams that animate on scroll. Cue jokes about turning the timeline into the “AI Netflix intro.”
Then came the Dark Mode Civil War. jvillasante begged for a light theme — “some people can’t read dark!” — while others swore the night mode saves their retinas. The spiciest thread? varispeed wondering if code‑writing AI like Codex intentionally “cosplays bugs,” forcing you to iterate like a human. Fans argued it’s natural trial‑and‑error; skeptics called it needless chaos. hmokiguess asked for charts, cycles, and a crystal‑ball prediction of the next boom. Between iconic stops like BERT and ChatGPT, the crowd demanded less lore, more pictures, and a forecast — turning a sober timeline into pure comment‑section cinema.
Key Points
- •The timeline traces major LLMs from 2017’s Transformer to 2022’s ChatGPT, highlighting architectures, training methods, and scale.
- •Key 2018–2019 milestones include ELMo, GPT-1, BERT, GPT-2, XLNet, RoBERTa, and T5, each advancing contextualization, pretraining, and task framing.
- •2020–2021 entries feature GPT-3’s few-shot capabilities, GShard’s MoE scaling, Switch Transformer’s sparse routing, open models by EleutherAI, and large dense models like Megatron-Turing NLG.
- •2022 emphasizes alignment and scaling insights: InstructGPT introduced RLHF, Chinchilla recalibrated scaling laws, and multiple open or large models (PaLM, OPT, BLOOM) emerged.
- •ChatGPT (GPT-3.5 with RLHF) popularized conversational AI, rapidly reaching 100M users and marking mainstream adoption.