September 30, 2025 at 1:20 PM
i think i’ve soured on diffusion language models
another one landed, scaled up to 30b-a3b, and it’s not *that* impressive. i don’t think we’re seeing scaling gains
www.radicalnumerics.ai/blog/rnd1
another one landed, scaled up to 30b-a3b, and it’s not *that* impressive. i don’t think we’re seeing scaling gains
www.radicalnumerics.ai/blog/rnd1
October 10, 2025 at 12:45 PM
i think i’ve soured on diffusion language models
another one landed, scaled up to 30b-a3b, and it’s not *that* impressive. i don’t think we’re seeing scaling gains
www.radicalnumerics.ai/blog/rnd1
another one landed, scaled up to 30b-a3b, and it’s not *that* impressive. i don’t think we’re seeing scaling gains
www.radicalnumerics.ai/blog/rnd1
sigh ok it probably is a llada variant, the founder has coyfaced "diffusion" pointers all over his comms. bitter lesson learner syndromed again.
September 16, 2026 at 12:51 PM
sigh ok it probably is a llada variant, the founder has coyfaced "diffusion" pointers all over his comms. bitter lesson learner syndromed again.
Alibaba (Ant Group)'s LLaDA-8B, a diffusion model with an unprecedented 8B scale
LLaDA is a masked diffusion model that follows standard pretraining and SFT while sampling via diffusion. During pretraining, it masks all tokens randomly at ratio; in SFT, only response tokens may be masked.
LLaDA is a masked diffusion model that follows standard pretraining and SFT while sampling via diffusion. During pretraining, it masks all tokens randomly at ratio; in SFT, only response tokens may be masked.
February 17, 2025 at 6:30 AM
Alibaba (Ant Group)'s LLaDA-8B, a diffusion model with an unprecedented 8B scale
LLaDA is a masked diffusion model that follows standard pretraining and SFT while sampling via diffusion. During pretraining, it masks all tokens randomly at ratio; in SFT, only response tokens may be masked.
LLaDA is a masked diffusion model that follows standard pretraining and SFT while sampling via diffusion. During pretraining, it masks all tokens randomly at ratio; in SFT, only response tokens may be masked.
Congratulations to GM Matthias Blübaum on winning the European #Chess Championship for a record second time after 2022! 🏆
Congrats also to all the 20 players who qualified for the upcoming FIDE World Cup! 🎉
lichess.org/broadcast/eu...
📷: David Llada / Federația Română de Șah / ECU
Congrats also to all the 20 players who qualified for the upcoming FIDE World Cup! 🎉
lichess.org/broadcast/eu...
📷: David Llada / Federația Română de Șah / ECU
March 26, 2025 at 7:06 PM
Congratulations to GM Matthias Blübaum on winning the European #Chess Championship for a record second time after 2022! 🏆
Congrats also to all the 20 players who qualified for the upcoming FIDE World Cup! 🎉
lichess.org/broadcast/eu...
📷: David Llada / Federația Română de Șah / ECU
Congrats also to all the 20 players who qualified for the upcoming FIDE World Cup! 🎉
lichess.org/broadcast/eu...
📷: David Llada / Federația Română de Șah / ECU
Large Language Diffusion Models
A wildly new AI architecture, this uses diffusion (all tokens at once), not next token prediction
ml-gsai.github.io/LLaDA-demo/
A wildly new AI architecture, this uses diffusion (all tokens at once), not next token prediction
ml-gsai.github.io/LLaDA-demo/
SOCIAL MEDIA TITLE TAG
SOCIAL MEDIA DESCRIPTION TAG TAG
ml-gsai.github.io
February 17, 2025 at 1:39 PM
Large Language Diffusion Models
A wildly new AI architecture, this uses diffusion (all tokens at once), not next token prediction
ml-gsai.github.io/LLaDA-demo/
A wildly new AI architecture, this uses diffusion (all tokens at once), not next token prediction
ml-gsai.github.io/LLaDA-demo/
there's a simple way of improving the sampling strategy of llada:
July 11, 2025 at 3:33 PM
there's a simple way of improving the sampling strategy of llada:
It looks like TypeSafe forked LLaDa and vLLM. No clue if they built Jev on LLaDa but definitely seems like they’re interested in diffusion at least.
github.com/typesafe-ai/...
github.com/typesafe-ai/...
GitHub - typesafe-ai/LLaDA: Official PyTorch implementation for "Large Language Diffusion Models"
Official PyTorch implementation for "Large Language Diffusion Models" - typesafe-ai/LLaDA
github.com
September 16, 2026 at 1:39 AM
It looks like TypeSafe forked LLaDa and vLLM. No clue if they built Jev on LLaDa but definitely seems like they’re interested in diffusion at least.
github.com/typesafe-ai/...
github.com/typesafe-ai/...
🚀 LLaDA: Diffusion Models for LLMs challenges the dominance of autoregressive models in LLMs! LLaDA, a diffusion-based Transformer, achieves strong in-context learning, competitive instruction-following, and even beats GPT-4o in reversal tasks. Trained with PyTorch Lightning ⚡
March 14, 2025 at 12:03 PM
🚀 LLaDA: Diffusion Models for LLMs challenges the dominance of autoregressive models in LLMs! LLaDA, a diffusion-based Transformer, achieves strong in-context learning, competitive instruction-following, and even beats GPT-4o in reversal tasks. Trained with PyTorch Lightning ⚡
LLaDA - Large Language Diffusion Models
arxiv.org/abs/2502.09992
Significant progress towards language diffusion models. Reportedly on par with LLaMA3 on many benchmarks.
arxiv.org/abs/2502.09992
Significant progress towards language diffusion models. Reportedly on par with LLaMA3 on many benchmarks.
March 1, 2025 at 10:03 PM
LLaDA - Large Language Diffusion Models
arxiv.org/abs/2502.09992
Significant progress towards language diffusion models. Reportedly on par with LLaMA3 on many benchmarks.
arxiv.org/abs/2502.09992
Significant progress towards language diffusion models. Reportedly on par with LLaMA3 on many benchmarks.
It’s still early days, but projects like LLaDA 8B (Feb 2025) are starting to compete with autoregressive models of similar size (LLaMA3).
Perhaps more importantly than just a compute advantage, LLaDA handles reversal-tests (right-to-left thinking) extremely well!
Perhaps more importantly than just a compute advantage, LLaDA handles reversal-tests (right-to-left thinking) extremely well!
May 5, 2025 at 7:26 PM
It’s still early days, but projects like LLaDA 8B (Feb 2025) are starting to compete with autoregressive models of similar size (LLaMA3).
Perhaps more importantly than just a compute advantage, LLaDA handles reversal-tests (right-to-left thinking) extremely well!
Perhaps more importantly than just a compute advantage, LLaDA handles reversal-tests (right-to-left thinking) extremely well!
LLMs That Don't Gaslight You
A new language model uses diffusion instead of next-token prediction. That means the text it can back out of a hallucination before it commits. This is a big win for areas like law & contracts, where global consistency is valued
timkellogg.me/blog/2025/02...
A new language model uses diffusion instead of next-token prediction. That means the text it can back out of a hallucination before it commits. This is a big win for areas like law & contracts, where global consistency is valued
timkellogg.me/blog/2025/02...
LLaDA: Large Language Diffusion Models
timkellogg.me
February 17, 2025 at 11:32 PM
LLMs That Don't Gaslight You
A new language model uses diffusion instead of next-token prediction. That means the text it can back out of a hallucination before it commits. This is a big win for areas like law & contracts, where global consistency is valued
timkellogg.me/blog/2025/02...
A new language model uses diffusion instead of next-token prediction. That means the text it can back out of a hallucination before it commits. This is a big win for areas like law & contracts, where global consistency is valued
timkellogg.me/blog/2025/02...
Tiene la muy turística localidad castrillonense de #SantaMaríadelMar dos hermosas playas cuando la marea sube. La del este, de La #Llada, en la desembocadura del arroyo de La #Ferrería, se presenta en la fotografía. Ambas son de finas arenas, seguras, con buenos accesos, duchas, vigilancia…
March 23, 2025 at 10:08 AM
Tiene la muy turística localidad castrillonense de #SantaMaríadelMar dos hermosas playas cuando la marea sube. La del este, de La #Llada, en la desembocadura del arroyo de La #Ferrería, se presenta en la fotografía. Ambas son de finas arenas, seguras, con buenos accesos, duchas, vigilancia…
Batalla entre La Revuelta y El intermedio. La berrea del ciervo.
Ángel Cristo Junior cuenta que tras sacar las fotos de su madre, Bárbara Rey, con Juan Carlos I, “me fui a casa de mi vecina Jenny Llada y me puse a ver Los caballeros del zodiaco, mis dibujos animados favoritos”.
Ángel Cristo Junior cuenta que tras sacar las fotos de su madre, Bárbara Rey, con Juan Carlos I, “me fui a casa de mi vecina Jenny Llada y me puse a ver Los caballeros del zodiaco, mis dibujos animados favoritos”.
December 20, 2024 at 10:11 AM
Batalla entre La Revuelta y El intermedio. La berrea del ciervo.
Ángel Cristo Junior cuenta que tras sacar las fotos de su madre, Bárbara Rey, con Juan Carlos I, “me fui a casa de mi vecina Jenny Llada y me puse a ver Los caballeros del zodiaco, mis dibujos animados favoritos”.
Ángel Cristo Junior cuenta que tras sacar las fotos de su madre, Bárbara Rey, con Juan Carlos I, “me fui a casa de mi vecina Jenny Llada y me puse a ver Los caballeros del zodiaco, mis dibujos animados favoritos”.
Teodora Injac is the only top women's player competing in the Singapore Open at the World Championship w/ 37 GMs!
After Round 4, she's co-leading the tournament w/ 7 others! Today she beat Singapore's No. 2 GM Siddharth Jagadeesh (2510)!
chess-results.com/tnr989481.as...
📷: D Llada
#womeninchess
After Round 4, she's co-leading the tournament w/ 7 others! Today she beat Singapore's No. 2 GM Siddharth Jagadeesh (2510)!
chess-results.com/tnr989481.as...
📷: D Llada
#womeninchess
December 1, 2024 at 4:09 PM
Teodora Injac is the only top women's player competing in the Singapore Open at the World Championship w/ 37 GMs!
After Round 4, she's co-leading the tournament w/ 7 others! Today she beat Singapore's No. 2 GM Siddharth Jagadeesh (2510)!
chess-results.com/tnr989481.as...
📷: D Llada
#womeninchess
After Round 4, she's co-leading the tournament w/ 7 others! Today she beat Singapore's No. 2 GM Siddharth Jagadeesh (2510)!
chess-results.com/tnr989481.as...
📷: D Llada
#womeninchess
La f*llada de boca que te quiero pegar. Buenos días!
August 2, 2025 at 6:24 AM
La f*llada de boca que te quiero pegar. Buenos días!
LLaDA 2.1 is out 🔥 MoE diffusion language models released by AntGroup
huggingface.co/inclusionAI/...
huggingface.co/inclusionAI/...
✨LLaDA2.1-mini: 16B - Apache2.0
✨LLaDA2.1-flash: 100B - Apache2.0
✨Both delivers editable generation, RL-trained diffusion reasoning and fast inference
huggingface.co/inclusionAI/...
huggingface.co/inclusionAI/...
✨LLaDA2.1-mini: 16B - Apache2.0
✨LLaDA2.1-flash: 100B - Apache2.0
✨Both delivers editable generation, RL-trained diffusion reasoning and fast inference
February 9, 2026 at 9:40 PM
LLaDA 2.1 is out 🔥 MoE diffusion language models released by AntGroup
huggingface.co/inclusionAI/...
huggingface.co/inclusionAI/...
✨LLaDA2.1-mini: 16B - Apache2.0
✨LLaDA2.1-flash: 100B - Apache2.0
✨Both delivers editable generation, RL-trained diffusion reasoning and fast inference
huggingface.co/inclusionAI/...
huggingface.co/inclusionAI/...
✨LLaDA2.1-mini: 16B - Apache2.0
✨LLaDA2.1-flash: 100B - Apache2.0
✨Both delivers editable generation, RL-trained diffusion reasoning and fast inference
this seems like reasonable analysis and actually agrees with my initial guess about what Jev was. but a bunch of people were saying how there's evidence typesafe forked llada and used that, which pushes more towards the diffusion direction.
did you have thoughts on that?
github.com/typesafe-ai/...
did you have thoughts on that?
github.com/typesafe-ai/...
October 2, 2026 at 7:41 PM
this seems like reasonable analysis and actually agrees with my initial guess about what Jev was. but a bunch of people were saying how there's evidence typesafe forked llada and used that, which pushes more towards the diffusion direction.
did you have thoughts on that?
github.com/typesafe-ai/...
did you have thoughts on that?
github.com/typesafe-ai/...
LLaDA: The Diffusion Model That Could Redefine Language Generation
How LLaDA works, why it matters, and how it could shape the next generation of LLMs
#ai #diffusionmodel #llm
How LLaDA works, why it matters, and how it could shape the next generation of LLMs
#ai #diffusionmodel #llm
LLaDA: The Diffusion Model That Could Redefine Language Generation
How LLaDA works, why it matters, and how it could shape the next generation of LLMs
towardsdatascience.com
February 28, 2025 at 1:04 AM
LLaDA: The Diffusion Model That Could Redefine Language Generation
How LLaDA works, why it matters, and how it could shape the next generation of LLMs
#ai #diffusionmodel #llm
How LLaDA works, why it matters, and how it could shape the next generation of LLMs
#ai #diffusionmodel #llm
reminds me of LLaDA, where they explained that autoregression is equal to diffusion, thus you absolutely can do diffusion for text.
maybe another relationship to add to the table?
timkellogg.me/blog/2025/02...
maybe another relationship to add to the table?
timkellogg.me/blog/2025/02...
LLaDA: LLMs That Don't Gaslight You
A new language model uses diffusion instead of next-token prediction. That means the text it can back out of a hallucination before it commits. This is a big win for areas like law & contracts, where ...
timkellogg.me
April 27, 2025 at 11:20 AM
reminds me of LLaDA, where they explained that autoregression is equal to diffusion, thus you absolutely can do diffusion for text.
maybe another relationship to add to the table?
timkellogg.me/blog/2025/02...
maybe another relationship to add to the table?
timkellogg.me/blog/2025/02...
La turística localidad castrillonense de Santa María del Mar (Asturias) tiene dos playas en pleamar:
Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
June 20, 2026 at 10:25 AM
La turística localidad castrillonense de Santa María del Mar (Asturias) tiene dos playas en pleamar:
Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
La turística localidad castrillonense de
#SantaMaríadelMar #Asturias tiene dos playas en pleamar: Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
#SantaMaríadelMar #Asturias tiene dos playas en pleamar: Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
May 23, 2026 at 9:38 AM
La turística localidad castrillonense de
#SantaMaríadelMar #Asturias tiene dos playas en pleamar: Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
#SantaMaríadelMar #Asturias tiene dos playas en pleamar: Esta, al poniente, igual que la de La Llada, reúne excelentes condiciones naturales, de seguridad, cuidado entorno, buenos accesos y servicios.
Acabo de pegarme la f*llada padre con mi colega el trípode canadiense que hemos muerto y resucitao varias veces, chavales...
a red teddy bear laying in the grass with the words * revives * written on it
ALT: a red teddy bear laying in the grass with the words * revives * written on it
media.tenor.com
April 5, 2026 at 10:47 AM
Acabo de pegarme la f*llada padre con mi colega el trípode canadiense que hemos muerto y resucitao varias veces, chavales...
Qué viva Canadá con sus grizzlies y sus guardabosques, la f*llada que nos hemos pegao, chavales, ay...
a group of cartoon characters are sitting at desks and one of them is wearing a hat that says canada
ALT: a group of cartoon characters are sitting at desks and one of them is wearing a hat that says canada
media.tenor.com
August 9, 2025 at 2:34 PM
Qué viva Canadá con sus grizzlies y sus guardabosques, la f*llada que nos hemos pegao, chavales, ay...