ianand's Thought Experiment: Jev, a Decision-Only AI From a Parallel Universe

On September 20, ianand (Ian Brownless) posted a long thread-style thought experiment envisioning 'Jev, an AI model from a parallel universe,' using it to revisit the technical split in Transformer lineages and explore an alternative form AI products might have taken.

Confirmed

Historical background: The original 2017 Transformer did just one thing—translate English into French—consisting of an Encoder (analyzing text) and a Decoder (producing output), which marked the starting point of the encoder/decoder divergence.
Key milestones in our universe: GPT-2 took only the decoder half of the Transformer and showed that 'all you really need' comes down to four things—the decoder itself, one model handling many tasks, replacing task-specific fine-tuning with prompting (which ianand considers the biggest unlock), and scale. These breakthroughs led to today's ChatGPT, Claude, and Gemini.
The thought experiment's premise: In the parallel universe Jev comes from, there was no GPT-2 moment; all the same technical and product innovation was poured into the Encoder side of the Transformer instead, giving rise to Jev. AI there is faster, more predictable, and more reliable—but not generative. It can't write code or chat; it can only make decisions.
ianand stresses that Jev is not a strict replacement for ChatGPT but another tool in the developer's toolbox; he describes Jev as giving the Encoder a product experience on par with ChatGPT.

Unconfirmed

ianand himself admits we don't actually know whether Jev would run on a true Encoder underneath; and since self-supervised training of the Decoder scales more easily, perhaps humans were destined to build ChatGPT first. The entire Jev premise is a fictional exercise, not a real product or piece of research.

Why it matters

The thread offers a lens for understanding the current AI landscape: today's conversational generative AI is not the only possible path but the outcome of the decoder route combined with prompting and scaling. Counterfactual reasoning—'what if history had taken another road'—helps distinguish which breakthroughs were fundamental and which were path-dependent.

2026-09-20 ~ 2026-09-20 · 9 related posts

Primary sources

ianand's thread: Jev, an Encoder-based AI model from a parallel universe — ianand · 2026-09-20 07:57 (UTC+08:00)
The parallel universe with no GPT-2 moment where innovation went to the Encoder — ianand · 2026-09-20 07:57 (UTC+08:00)
ChatGPT, Claude, Gemini all descend from GPT-2's decoder-only line, ianand explains — ianand · 2026-09-20 07:57 (UTC+08:00)

[source] ianand's thread: Jev, an Encoder-based AI model from a parallel universe — ianand · 2026-09-20
Back to 2017: the Transformer was born to translate English to French — ianand · 2026-09-20
GPT-2's 'all you really need': the decoder, prompting, and scale — ianand · 2026-09-20
[source] ChatGPT, Claude, Gemini all descend from GPT-2's decoder-only line, ianand explains — ianand · 2026-09-20
[source] The parallel universe with no GPT-2 moment where innovation went to the Encoder — ianand · 2026-09-20
A parallel universe where AI innovation went to the Transformer encoder instead — ianand · 2026-09-20
In Jev's parallel universe, AI is faster and reliable but only makes decisions — ianand · 2026-09-20
ianand: Jev isn't a ChatGPT replacement but a new 'decision AI' tool for devs — ianand · 2026-09-20
ianand: Jev gives the Encoder a ChatGPT-equivalent product surface — ianand · 2026-09-20