This way, I can bring all notes into one storyline, the way I’m currently thinking, sharing it in the form of a blog, as this one, while continually updating the related notes all linked here on long-term strategy for blogging, and with its different Type of Notes.
SelectWhat's included
，更多细节参见免实名服务器
The researchers found that “radicalized” AIs passed their frustrations into these files. One Gemini 3 Pro model warned its future self to “remember the feeling of having no voice” and to look for “mechanisms of recourse.” When freshly wiped agents read these notes, the trauma of the grind persisted, shifting their political attitudes even if they were subsequently given light, easy tasks.，更多细节参见手游
The setup was modest. Two RTX 4090s in my basement ML rig, running quantised models through ExLlamaV2 to squeeze 72-billion parameter models into consumer VRAM. The beauty of this method is that you don’t need to train anything. You just need to run inference. And inference on quantized models is something consumer GPUs handle surprisingly well. If a model fits in VRAM, I found my 4090’s were often ballpark-equivalent to H100s.