Offlined v3.0.0 is here.
The biggest update since it was born: courses, a 3D universe, the seafloor of the whole planet, 3D maps, family profiles and a full security audit.
No internet. No middlemen.
https://t.co/mDj1oAVRET
@PiperUnplugged Fun episode idea. On a laptop you can go one step further: full Wikipedia plus offline maps and a local AI that answers from it (that's what we build at https://t.co/65e5ajbY3g). Which Wikipedia size did you go with on the phone?
@X_net_ Y gana aún más cuando le das fuentes propias: una IA local que consulta una Wikipedia guardada en tu disco y te dice de qué artículo sale cada dato. Conocimiento y datos, todo en casa.
@viniragazzi Boa lista. O mesmo vale para o computador: Wikipédia inteira, mapas e uma IA local que funcionam sem conexão. E o ponto principal vale ouro: tudo tem que ser baixado antes da crise.
@jiya_3063 Not dumb at all. It lives in the weights: a few GB of numbers that compress patterns from the training text. That's also why it gets fuzzy on details, and why giving it an offline Wikipedia to look things up helps so much.
@NixFred Love that it's for friends, not a product. One fun add-on: a Kiwix Wikipedia dump as a knowledge source in Open WebUI, so the smaller models cite instead of guess. Which model do your friends end up using most? ^^
@numeribrain Bonne nouvelle pour la vie privée, et pouvoir tout couper est un vrai plus. L'étape suivante : des modèles locaux qui s'appuient sur une Wikipédia hors ligne pour citer leurs sources au lieu d'inventer.
Offlined v3.0.0 is here.
The biggest update since it was born: courses, a 3D universe, the seafloor of the whole planet, 3D maps, family profiles and a full security audit.
No internet. No middlemen.
https://t.co/mDj1oAVRET
@SergioGMN Para mí eso es lo que quita el miedo: con una IA local, si mañana cambian precios o condiciones, tu herramienta sigue ahí y tus datos también. ¿Qué modelo usas en local para el día a día?
@JaicSam What a makerspace list. Offline Wikipedia + GPS maps is the combo we keep coming back to as well; add a small local model that reads the articles and it turns into a pocket librarian. ^^
@yabhishekhd 128GB of unified memory in a laptop is a big jump for local AI. The real test will be the software side. Which of these would you actually pick for running models day to day?
The wettest place on Earth gets almost 12 metres of rain a year. 🇮🇳☔
Mawsynram, in the hills of Meghalaya, averages 11,872 mm. In 1985 it got 26,000 mm, a Guinness record.
Found offline with Offlined's WikiNets.
https://t.co/65e5ajbY3g #DidYouKnow
@silver_speaker@joshuaherrerag@amigafreelance Justo, en eso te damos la razón. Con artículos de Wikipedia un 4B va bien, pero para PDFs mejor un 27B y una buena máquina, sin duda. Y lo que hemos visto es que en un PC doméstico normal el RAG todavía no compensa.
@Santiagov77@joshuaherrerag@amigafreelance Depende de qué llames decente. Para charlar y consultar cosas, un modelo de 4B corre en un PC normal de 16 GB. Y si le das una Wikipedia offline donde mirar, acierta mucho más sin máquina de 3.000 €.
@_nasch_ Y no solo para programar: para consultar tus documentos o una Wikipedia offline, un modelo pequeño en un PC normal ya va sobrado, sin cuotas de tokens y sin que nada salga de tu equipo.
@ariyahere0 Nice setup for an old laptop. One thing that helps a lot at that size: give the model an offline Wikipedia to look things up in, so it checks facts instead of guessing. That's the idea behind Offlined: https://t.co/65e5ajbqdI