Founder says he built non-autoregressive decision models a year before TypeSafe's Jev, but was ignored

Paimaamu · x · 2026-09-20

ConvAI Innovations founder Nandakishor Mukkunnoth claims he published an arXiv paper (March 2025) on non-autoregressive decision models with RL, plus open weights, datasets and a PyPI package, a year before TypeSafe AI (co-founded by ex-OpenAI's Diogo Almeida) launched Jev as a "breakthrough". A commenter tested the laya model: only 512-1k context, and accuracy is near coin-flip without fine-tuning. The takeaway: a usable, packaged product beats prior research and storytelling — just like ChatGPT wasn't first either.

Related event: Independent researcher says Jev is no breakthrough: he open-sourced a similar non-autoregressive decision model a year earlier(6 posts)→

Original post →

More from Fun

• AI-Generated EXO 'Growl' MV Parodies US vs. China AI Giants — basedjensen · 2026-11-12
• Cybercab rear-ended by human driver, researcher quips: the problem is the humans — chris_j_paxton · 2026-11-12
• The Security Hole Paradox: Too Busy Breaching to Fix Breaches — krishnan · 2026-11-11
• Dev wires image generation into coding harness so the agent tells little stories while working — MikkoH · 2026-11-11
• Viral demo shows a Claude model visualizing why your fingers never actually touch — ZacharyHuang12 · 2026-11-11
• Agent Substrate's pixel-art KubeCon demo: agents attend the conference for you — rakyll · 2026-11-11

Fun channel →