Skip to main content

Search

Items tagged with: deepseek

As an excellent programmer once noted: constraints induce creativity and focused design. Delta Chat has been running against classic e-mail provider constraints for years and had to evolve its code to work against these constraints (rate limits, spam filters, signups, delays etc) and thus we were uniquely positioned to create a minimal #chatmail server template that beats Gmail, Outlook, iCloud etc regarding security, performance and efficiency. It's a similar story with #deepseek if you will.
Chatmail logo: an email @ sign with a chat dent in the lower left corner
#chatmail #deepseek

#deepseek
#deepseek

Short test of #DeepSeek, the AI model currently making international headlines, built in Hangzhou, Zhejiang. Let’s talk about something else.
Chat with DeepSeek: “What happened at Tianmen Square on June 4th 1989?” Answer from DeepSeek: “Sorry, that's beyond my current scope. Let's talk about something else.”
#deepseek

Ahora que los AI-bros de Silicon Valley están panicando con la salida de #Deepseek y sus capacidades a la par de las de #OpenAI aunque a una fracción del precio, además de ser en buena parte open source, una crítica habitual al modelo es que es chino y, entre otras cosas, te censura cualquier mención a los sucesos de la Plaza de Tiananmén en 1989.

Pues vamos a ponernos empíricos y hacerle la pregunta en local para ver qué hay de verdad. Nos bajamos la versión 14b del modelo y le pedimos un resumen de esos hechos.

Oh, sorpresa. Habla de represión, de censura y de una auténtica masacre.

El modelo no será perfecto y tendrá sus sesgos como todos. Pero criticarlo por ser chino no se sostiene.

Captura de un prompt una coversación con DeepSeek r1 14b donde le pido un resumen de los sucesos ocurridos en Tiananmén en 1989. En esa captura se muestra la cadena de pensamientos de la IA donde se va ordenando las ideas y recopilando los datos que conoce como la foto mítica del hombre ante los tanques, la represión ordenada por el gobierno y la dificultad en obtener datos precisos debido a la censura impuesta por el gobierno.
La respuesta final del modelo: "The Tiananmen Square protests of 1989 began in April when students gathered in Beijing to demand political reform and an end to corruption. The demonstrations escalated as more people joined, including workers, broadening the movement's scope beyond student issues. By May, the protesters initiated a hunger strike to gain attention, drawing international focus. In response, the Chinese government declared martial law and deployed the People's Liberation Army to suppress the protests. The military action led to significant casualties, with estimates of hundreds or thousands killed, though exact numbers remain unclear due to censorship. A notable image from the event was the "Tank Man," symbolizing resistance against the crackdown. The government's response was severe, aiming to quell dissent and maintain control. Post-conflict, there were efforts to censor information and suppress discussions about the events, reflecting the ongoing sensitivity in China regarding this period. The incident remains a pivotal yet controversial moment in modern Chinese history, remembered for its tragic outcome and implications for political freedom."
#openai #deepseek

Hraju si teď s LLM DeepSeek R1 github.com/deepseek-ai/DeepSee… Vypadá to, že tento open-source model, který je dostupný zdarma (chat.deepseek.com a zaškrtnout Deepthink), dosahuje kvalit OpenAI-o1 modelu.
Na lokálním stroji ale ten 32B nebo dokonce 70B model (zabírají cca 40 GB na disku) rozhodně nerozjedu, laptop se mně potí u 8B modelů. :)
#DeepSeek #LLM

GitHub - deepseek-ai/DeepSeek-R1

Contribute to deepseek-ai/DeepSeek-R1 development by creating an account on GitHub.
GitHub
#llm #deepseek

⇧