#qwen — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #qwen, aggregated by home.social.
-
I've been working on my #raspberrypi powered #chess playing robot. I hit a wall trying to implement this design: https://www.instructables.com/Chess-Robot-Raspberry-Pi-Lynxmotion-AL5D-Arm/ This week I turned to #opencode and #qwen and have been astounded at how quickly it's managed to make board and piece detection so robust, spin up a ui to calibrate, change settings, track a game. I've been ambivalent about LLMs but its hard not to be astounded by their capacity. After the technical kinks get worked out I'll take the chainsaw to an #oak cutting I've had aging in the shed for 5 years and house the whole thing on a wooden table. The #pi will be hid in a hollow book.
-
I've been working on my #raspberrypi powered #chess playing robot. I hit a wall trying to implement this design: https://www.instructables.com/Chess-Robot-Raspberry-Pi-Lynxmotion-AL5D-Arm/ This week I turned to #opencode and #qwen and have been astounded at how quickly it's managed to make board and piece detection so robust, spin up a ui to calibrate, change settings, track a game. I've been ambivalent about LLMs but its hard not to be astounded by their capacity. After the technical kinks get worked out I'll take the chainsaw to an #oak cutting I've had aging in the shed for 5 years and house the whole thing on a wooden table. The #pi will be hid in a hollow book.
-
I've been working on my #raspberrypi powered #chess playing robot. I hit a wall trying to implement this design: https://www.instructables.com/Chess-Robot-Raspberry-Pi-Lynxmotion-AL5D-Arm/ This week I turned to #opencode and #qwen and have been astounded at how quickly it's managed to make board and piece detection so robust, spin up a ui to calibrate, change settings, track a game. I've been ambivalent about LLMs but its hard not to be astounded by their capacity. After the technical kinks get worked out I'll take the chainsaw to an #oak cutting I've had aging in the shed for 5 years and house the whole thing on a wooden table. The #pi will be hid in a hollow book.
-
I've been working on my #raspberrypi powered #chess playing robot. I hit a wall trying to implement this design: https://www.instructables.com/Chess-Robot-Raspberry-Pi-Lynxmotion-AL5D-Arm/ This week I turned to #opencode and #qwen and have been astounded at how quickly it's managed to make board and piece detection so robust, spin up a ui to calibrate, change settings, track a game. I've been ambivalent about LLMs but its hard not to be astounded by their capacity. After the technical kinks get worked out I'll take the chainsaw to an #oak cutting I've had aging in the shed for 5 years and house the whole thing on a wooden table. The #pi will be hid in a hollow book.
-
I've been working on my #raspberrypi powered #chess playing robot. I hit a wall trying to implement this design: https://www.instructables.com/Chess-Robot-Raspberry-Pi-Lynxmotion-AL5D-Arm/ This week I turned to #opencode and #qwen and have been astounded at how quickly it's managed to make board and piece detection so robust, spin up a ui to calibrate, change settings, track a game. I've been ambivalent about LLMs but its hard not to be astounded by their capacity. After the technical kinks get worked out I'll take the chainsaw to an #oak cutting I've had aging in the shed for 5 years and house the whole thing on a wooden table. The #pi will be hid in a hollow book.
-
Alibaba releases Qwen 3.8 Omni Flash
https://qwen.ai/blog?id=qwen3.8-omni-flash
Comments: https://news.ycombinator.com/item?id=49747925
#HackerNews #Alibaba #Qwen #OmniFlash #AI #technology #updates
-
Alibaba releases Qwen 3.8 Omni Flash
https://qwen.ai/blog?id=qwen3.8-omni-flash
Comments: https://news.ycombinator.com/item?id=49747925
#HackerNews #Alibaba #Qwen #OmniFlash #AI #technology #updates
-
Alibaba releases Qwen 3.8 Omni Flash
https://qwen.ai/blog?id=qwen3.8-omni-flash
Comments: https://news.ycombinator.com/item?id=49747925
#HackerNews #Alibaba #Qwen #OmniFlash #AI #technology #updates
-
Alibaba releases Qwen 3.8 Omni Flash
https://qwen.ai/blog?id=qwen3.8-omni-flash
Comments: https://news.ycombinator.com/item?id=49747925
#HackerNews #Alibaba #Qwen #OmniFlash #AI #technology #updates
-
Alibaba releases Qwen 3.8 Omni Flash
https://qwen.ai/blog?id=qwen3.8-omni-flash
Comments: https://news.ycombinator.com/item?id=49747925
#HackerNews #Alibaba #Qwen #OmniFlash #AI #technology #updates
-
Последний пеликан
Что-то много пеликанов на велосипеде начало появляться в ленте моего информационного пузыря в твиттер и я решил провести собственный эксперимент: GPT Astra против Qwen 3.8 и других доступных моделей. Цель: Охватить экспериментом повседневные модели, которые можно легко проверить на качество генерации и получить сравнительную оценку их способностей.
-
🧵 Suite de mes pérégrinations avec les LLM désinhibés…
Une étude intéressante publiée par la société 10a Lab il y a quelques jours, "Uncensored Open-weight Models: Redistribution as the Persistence Layer" ( https://arxiv.org/abs/2609.05241 ), a cartographié l’écosystème Hugging Face/GitHub autour de ces modèles.
Résultat : 3 471 modèles uncensored originaux, déjà transformés en 8 164 redistributions quantifiées/repackagées.
Et surtout : sur 1 643 applications GitHub utilisant ce type de backend, 411, donc 25%, sont classées par 10a Labs comme explicitement comme malveillantes : hacking, fraude, génération de malware, etc.
Autre détail qui résonne pas mal avec mes petits tests locaux : les auteurs constatent que la quantification et le repackaging rendent ces modèles facilement déployables sur du matériel grand public.
Les modèles 3–8B constituent d’ailleurs 41 % de leur dataset.Et Qwen est le king dans ce paysage : parmi les modèles d’origine chinoise identifiés, 80 % sont basés sur la famille Qwen.
"Cerise sur le gâteau" : avec Heretic, retirer les garde-fous est passé d’un workflow nécessitant une certaine maîtrise des entrailles… à quasiment une commande terminal.
Dans leur dataset, la production est passée d’environ 89 modèles/mois avant Heretic à ~338/mois après.Et pendant ce temps, Check Point retrouve dans les échanges d'un groupe cybercriminel ransomware la recommandation suivante:
“Qwen 3.5 with all barriers removed… Zero refusals. Absolutely no restrictions.”
( https://engage.checkpoint.com/ai-security-report-2026/items/report )
Reste donc une question qui devient de moins en moins théorique :
Est-ce qu’on va voir rapidement une bascule vers du “bullet-proof self-hosting”, hors télémétrie, hors contrôle fournisseur, ou est-ce que le vol de comptes, de clés API et les marchés gris de tokens resteront encore longtemps la voie de moindre résistance pour les cybercriminels ?