Researcher claims TypeSafe AI's Jev repackages his year-old non-autoregressive model concept
An open-source AI researcher alleges a well-funded startup launched the same architecture he published a year earlier as a novel breakthrough.
What to know
- An independent researcher published non-autoregressive decision models and their RL-based framework in March and September 2025, releasing code, datasets, and papers openly.
- TypeSafe AI launched Jev in September 2026, a commercial product using the same core architecture but closed-source, without technical papers or open training data.
- The researcher publicly alleges concept replication and countered with RL Agent, claiming it is 4x faster and fully open-source.
Nandakishor M Independent AI researcherDiogo Almeida TypeSafe AI founder, ChatGPT co-inventorTypeSafe AI Commercial AI startup
How it unfolded 1 development · click the chart to see its coverage posts
-
1
Nandakishor M releases RL Agent, an open-source alternative
Rather than remain bitter, the researcher built RL Agent, a completely open-source System 1 decision model using a bidirectional encoder. He claims it runs in 33–38 milliseconds on GPU, approximately 4x faster than Jev's 150 ms latency.
“And because we built it properly on a bidirectional encoder, our model runs in 33 to 38 milliseconds on a GPU, making it roughly 4x faster than Jev's published 150 ms latency, and it is 100% open-source.”
— Nandakishor M -
> In my March 2025 work, I used frozen sequence embeddings combined with a separate PPO value network. It worked for turn-by-turn sales prediction, but it was not end-to-end **and could not handle dynamic new questions at runtime**. (emphasis mine) Maybe this is just my lack of ML knowledge showing, but isn’t this kind of the entire selling point…
2 more of the top 3 · 7 posts in this stretch
-
B
🔬 The best bytes of # science & # tech across the # fediverse “I Built Non-Autoregressive Decision Models a Year Ago․ Then a Frontier Lab Called It a 'Breakthrough' dev․to/nandakishor_m_6cc0adfde9f/i-built-non-autoregressive-decision-models-a-year-ago-then-a-frontier-lab-called-it…” https:// dev.to/nandakishor_m_6cc0adfde…
-
I can be wrong, but my impression is that they are building a generic classifier, for which you supply the task in a natural language and the schema of the output, and get some (hopefully, good) predictions. Classifiers are obviously not a new thing, but the old way is to gather a labeled dataset specifically for your task and then train a model…
-
-
background
Nandakishor M publicly accuses TypeSafe AI of repackaging his work — The researcher posted a detailed account on Dev.to (cross-posted to Lobsters), arguing that Jev copied the core concept he had published a year earlier and alleging the funded lab received credit for the idea while his open-source work went overlooked due to its narrower initial application.
-
background
TypeSafe AI launches Jev commercial product — TypeSafe AI, founded by ChatGPT co-inventor Diogo Almeida, launched Jev, a non-autoregressive decision model product using reinforcement learning for calibrated decisions (RLCD), priced at $0.042 per million input tokens with 150 ms response times. No technical papers, open weights, or training datasets were released.
-
background
Researcher publishes framework for reinforcement learning-guided decisions — Nandakishor M published a second arXiv paper (arXiv:2510.01237) laying out a framework for schema-based decisions guided by reinforcement learning, intended to generalize the earlier vertical sales application.
-
background
Nandakishor M publishes non-autoregressive decision model research — Researcher published an arXiv paper (arXiv:2503.23303) describing a non-autoregressive architecture for schema-based decisions, released model weights on Hugging Face, published an open dataset, and built a PyPi package.
What people are saying 4 voices from 1 site · best of 7 · verbatim
- Sep 20
-
Their original approach had that limitation. It looks like the new one (continuing with lessons learnt from 2025) supports any questions.
-
It's both a pro and a con for Jev. It contains lots of world knowledge so you can go more high-level. On the other hand you can't tune it to your needs too much. If you can't sufficiently describe the rules you need and potentially override existing ideas, you won't benefit from Jev.
- Sep 19
-
Should be merged with https://lobste.rs/s/ojukrw/laya_33ms_multilingual_system_1_decision, I think
-
I haven’t been in the trade for 20 years but, in my days, I solved the "decision problem" with simple bayesian learning. When the input is text to classify and the number of classification is low, bayesian are really really good.