-
@ lemon
2025-02-22 04:18:48
I don’t know about full on back doors in DeepSeek, but I am skeptical about the intentions of those behind Language Models in general
Each model is biased on the training data it was given so each model will be embedded with a culture, political ideology, and religious reasoning that will seed each user interaction
This is true for DeepSeek, Gemini, Grok, OpenAI and every other model
Ask DeepSeek about Tiananmen Square or Gemini about misgendering to prevent nuclear war
AI models could be the next advancement in digital colonialism that can infiltrate and influence the mind of the user
Each prompt can be a nudge akin to a TikTok, Tweet or Instagram Reel
Fortunately there’s competition and thankfully there’s Nostr to serve a future training ground for models that’s outside of political censorship or algorithmic bias
This leaves me optimistic