fuck you *re-autoregressivizes your non-autoregressive model*
September 27, 2026 at 7:07 PM
fuck you *re-autoregressivizes your non-autoregressive model*
This seems to be the most popular alternative to jev.
Laya: Multilingual, non-autoregressive System 1 decision model.
huggingface.co/convaiinnova...
Laya Repo: github.com/NandhaKishor...
Laya MLX: huggingface.co/aac6fef/laya...
Laya MLX Repo: github.com/mizorewww/la...
Laya: Multilingual, non-autoregressive System 1 decision model.
huggingface.co/convaiinnova...
Laya Repo: github.com/NandhaKishor...
Laya MLX: huggingface.co/aac6fef/laya...
Laya MLX Repo: github.com/mizorewww/la...
September 22, 2026 at 5:29 AM
This seems to be the most popular alternative to jev.
Laya: Multilingual, non-autoregressive System 1 decision model.
huggingface.co/convaiinnova...
Laya Repo: github.com/NandhaKishor...
Laya MLX: huggingface.co/aac6fef/laya...
Laya MLX Repo: github.com/mizorewww/la...
Laya: Multilingual, non-autoregressive System 1 decision model.
huggingface.co/convaiinnova...
Laya Repo: github.com/NandhaKishor...
Laya MLX: huggingface.co/aac6fef/laya...
Laya MLX Repo: github.com/mizorewww/la...
Well this is exciting: arxiv.org/abs/2512.20605
Emergent temporal abstractions in autoregressive models enable hierarchical reinforcement learning
Large-scale autoregressive models pretrained on next-token prediction and finetuned with reinforcement learning (RL) have achieved unprecedented success on many problem domains. During RL, these model...
arxiv.org
January 6, 2026 at 7:53 PM
Well this is exciting: arxiv.org/abs/2512.20605
A new research states that in a data-constrained environment, Diffusion models outperforms Autoregressive models.
Across different unique data scales, they observe:
1️⃣ At low compute, Autoregressive models win.
Across different unique data scales, they observe:
1️⃣ At low compute, Autoregressive models win.
July 22, 2025 at 11:09 PM
A new research states that in a data-constrained environment, Diffusion models outperforms Autoregressive models.
Across different unique data scales, they observe:
1️⃣ At low compute, Autoregressive models win.
Across different unique data scales, they observe:
1️⃣ At low compute, Autoregressive models win.
#PAISS2025 continues with Yann LeCun talking about world models and why generative autoregressive models are not the future.
September 4, 2025 at 9:44 AM
#PAISS2025 continues with Yann LeCun talking about world models and why generative autoregressive models are not the future.
lessons from an autoregressive being
July 16, 2026 at 2:53 PM
lessons from an autoregressive being
new concept: AI license. you can only use it if you can explain how autoregressive generation and attention work. that should cut down on bullshit. regulate it like ham radio
April 22, 2026 at 7:13 AM
new concept: AI license. you can only use it if you can explain how autoregressive generation and attention work. that should cut down on bullshit. regulate it like ham radio
an industry secret is that you can make up any old shit about LLMs and say it's "because they're autoregressive" and ppl will just believe you
April 7, 2026 at 4:18 PM
an industry secret is that you can make up any old shit about LLMs and say it's "because they're autoregressive" and ppl will just believe you
The return of the Autoregressive Image Model: AIMv2 now going multimodal.
Excellent work by @alaaelnouby.bsky.social & team with code and checkpoints already up:
arxiv.org/abs/2411.14402
Excellent work by @alaaelnouby.bsky.social & team with code and checkpoints already up:
arxiv.org/abs/2411.14402
November 22, 2024 at 9:44 AM
The return of the Autoregressive Image Model: AIMv2 now going multimodal.
Excellent work by @alaaelnouby.bsky.social & team with code and checkpoints already up:
arxiv.org/abs/2411.14402
Excellent work by @alaaelnouby.bsky.social & team with code and checkpoints already up:
arxiv.org/abs/2411.14402
🚨Model Checking for Vector Autoregressive Models 🚨
In a new preprint, @joranjongerling.bsky.social, @bsiepe.bsky.social, @sachaepskamp.bsky.social, Lourens Waldorp and I provide a tutorial on model checking for Vector Autoregressive (VAR) models: osf.io/preprints/ps...
In a new preprint, @joranjongerling.bsky.social, @bsiepe.bsky.social, @sachaepskamp.bsky.social, Lourens Waldorp and I provide a tutorial on model checking for Vector Autoregressive (VAR) models: osf.io/preprints/ps...
December 3, 2025 at 6:52 AM
🚨Model Checking for Vector Autoregressive Models 🚨
In a new preprint, @joranjongerling.bsky.social, @bsiepe.bsky.social, @sachaepskamp.bsky.social, Lourens Waldorp and I provide a tutorial on model checking for Vector Autoregressive (VAR) models: osf.io/preprints/ps...
In a new preprint, @joranjongerling.bsky.social, @bsiepe.bsky.social, @sachaepskamp.bsky.social, Lourens Waldorp and I provide a tutorial on model checking for Vector Autoregressive (VAR) models: osf.io/preprints/ps...
watching 4o make images, it doesn't look like it's doing normal tiled autoregressive image synthesis. i wonder if it's doing wavelet autoregression.
March 29, 2025 at 2:06 AM
watching 4o make images, it doesn't look like it's doing normal tiled autoregressive image synthesis. i wonder if it's doing wavelet autoregression.
Have you ever wondered how to train an autoregressive generative transformer on text and raw pixels, without a pretrained visual tokenizer (e.g. VQ-VAE)?
We have been pondering this during summer and developed a new model: JetFormer 🌊🤖
arxiv.org/abs/2411.19722
A thread 👇
1/
We have been pondering this during summer and developed a new model: JetFormer 🌊🤖
arxiv.org/abs/2411.19722
A thread 👇
1/
December 2, 2024 at 4:41 PM
Have you ever wondered how to train an autoregressive generative transformer on text and raw pixels, without a pretrained visual tokenizer (e.g. VQ-VAE)?
We have been pondering this during summer and developed a new model: JetFormer 🌊🤖
arxiv.org/abs/2411.19722
A thread 👇
1/
We have been pondering this during summer and developed a new model: JetFormer 🌊🤖
arxiv.org/abs/2411.19722
A thread 👇
1/
Simply masking 15% of input tokens + next-token prediction (NTP) can significantly boost LLMs on key information retrieval & long-context reasoning—without extra compute!
MEAP (Mask-Enhanced Autoregressive Prediction)
MEAP (Mask-Enhanced Autoregressive Prediction)
February 16, 2025 at 11:30 PM
Simply masking 15% of input tokens + next-token prediction (NTP) can significantly boost LLMs on key information retrieval & long-context reasoning—without extra compute!
MEAP (Mask-Enhanced Autoregressive Prediction)
MEAP (Mask-Enhanced Autoregressive Prediction)
One thing I’ve struggled with talking about Jev is whether or not it’s a *generative* model. It does generate something new, but in a much different and more limited sense than an autoregressive language model
September 26, 2026 at 5:11 PM
One thing I’ve struggled with talking about Jev is whether or not it’s a *generative* model. It does generate something new, but in a much different and more limited sense than an autoregressive language model
LLMs hallucinate because they're autoregressive ✅
LLMs can't hallucinate because they're autoregressive ✅
LLMs can't hallucinate because they're autoregressive ✅
April 7, 2026 at 4:19 PM
LLMs hallucinate because they're autoregressive ✅
LLMs can't hallucinate because they're autoregressive ✅
LLMs can't hallucinate because they're autoregressive ✅
Magi-1: The Autoregressive Diffusion Video Generation Model
🥇 The first autoregressive video model with top-tier quality output
🔓 100% open-source & tech report
📊 Exceptional performance on major benchmarks
🥇 The first autoregressive video model with top-tier quality output
🔓 100% open-source & tech report
📊 Exceptional performance on major benchmarks
April 22, 2025 at 6:10 AM
Magi-1: The Autoregressive Diffusion Video Generation Model
🥇 The first autoregressive video model with top-tier quality output
🔓 100% open-source & tech report
📊 Exceptional performance on major benchmarks
🥇 The first autoregressive video model with top-tier quality output
🔓 100% open-source & tech report
📊 Exceptional performance on major benchmarks
LLMs aren't conscious because they're autoregressive ✅ yup
LLMs are conscious because they're autoregressive ✅ also yup
LLMs are conscious because they're autoregressive ✅ also yup
April 7, 2026 at 4:19 PM
LLMs aren't conscious because they're autoregressive ✅ yup
LLMs are conscious because they're autoregressive ✅ also yup
LLMs are conscious because they're autoregressive ✅ also yup
evident in the use of jev as an autoregressive model
September 22, 2026 at 11:14 PM
evident in the use of jev as an autoregressive model
doom renderer as a llm ood.dev/posts/doom/
Doom, compiled into a transformer — Out of Distribution
No training anywhere: a compiler sets the weights of a stock Phi-3-architecture checkpoint so autoregressive generation executes Doom's classic renderer.
ood.dev
August 14, 2026 at 9:11 AM
doom renderer as a llm ood.dev/posts/doom/
π-PrimeNovo: an accurate and efficient non-autoregressive deep learning model for de novo peptide sequencing www.nature.com/artic...
---
#proteomics #prot-paper
---
#proteomics #prot-paper
January 3, 2025 at 5:40 PM
π-PrimeNovo: an accurate and efficient non-autoregressive deep learning model for de novo peptide sequencing www.nature.com/artic...
---
#proteomics #prot-paper
---
#proteomics #prot-paper
[1/2] We've released the code for LegoGPT. Our autoregressive model generates physically stable and buildable designs from text prompts by integrating physics laws and assembly constraints into LLM training and inference.
Code: github.com/AvaLovelace1...
Website: avalovelace1.github.io/LegoGPT/
Code: github.com/AvaLovelace1...
Website: avalovelace1.github.io/LegoGPT/
May 10, 2025 at 3:06 AM
[1/2] We've released the code for LegoGPT. Our autoregressive model generates physically stable and buildable designs from text prompts by integrating physics laws and assembly constraints into LLM training and inference.
Code: github.com/AvaLovelace1...
Website: avalovelace1.github.io/LegoGPT/
Code: github.com/AvaLovelace1...
Website: avalovelace1.github.io/LegoGPT/
LLM alignment is easy because they're autoregressive ✅
LLM alignment is hard because they're autoregressive ✅
LLM alignment is hard because they're autoregressive ✅
April 7, 2026 at 4:20 PM
LLM alignment is easy because they're autoregressive ✅
LLM alignment is hard because they're autoregressive ✅
LLM alignment is hard because they're autoregressive ✅
METAGENE-1: a 7B parameter autoregressive transformer model trained on over 1.5T DNA and RNA base pairs sequenced from wastewater samples.
@oliu-io.bsky.social @samsja19.bsky.social @shangshang-wang.bsky.social @jefftk.com @willieneis.bsky.social
metagene.ai/metagene-1-p...
@oliu-io.bsky.social @samsja19.bsky.social @shangshang-wang.bsky.social @jefftk.com @willieneis.bsky.social
metagene.ai/metagene-1-p...
January 6, 2025 at 10:19 PM
METAGENE-1: a 7B parameter autoregressive transformer model trained on over 1.5T DNA and RNA base pairs sequenced from wastewater samples.
@oliu-io.bsky.social @samsja19.bsky.social @shangshang-wang.bsky.social @jefftk.com @willieneis.bsky.social
metagene.ai/metagene-1-p...
@oliu-io.bsky.social @samsja19.bsky.social @shangshang-wang.bsky.social @jefftk.com @willieneis.bsky.social
metagene.ai/metagene-1-p...
It’s an autoregressive transformer architecture, just like LLMs, it’s just trained in different data so try again
August 8, 2026 at 3:51 PM
It’s an autoregressive transformer architecture, just like LLMs, it’s just trained in different data so try again