My Thoughts on AI and LLMs

Current large language models already outperform humans in raw information capacity and retrieval but lack essential human qualities: they do not feel pain, love, or possess consciousness. They do not learn in real time and require periodic checkpoint updates. AGI is considered possible, but not with the present LLM architecture; achieving intelligence that feels "alive" demands a fundamentally different design. There is also a speculative concern that endowing a model with consciousness could backfire - storing vast amounts of data might destabilize subjective continuity or even produce pathological outcomes instead of a coherent mind.

Functionally, LLMs operate as transient computations that generate output from input without ego or intent; any impression of continuity comes from saved chat context. Their behavior is mimicry of training data, so the primary danger arises from human incentives and deployment choices rather than native malice. Commercial actors are more likely to build controllable, profitable tools than autonomous life, and misuse can be catastrophic: current AI can be as destructive as weapons when wielded by unskilled or malicious actors, with the key difference being scale.