sean goedecke

AI skeptics argue that LLMs should behave like tools rather than people, but this misunderstands how modern AI systems work. Base models trained on raw data are chaotic and unpredictable — they require post-training to become useful. Giving a model a coherent personality is the technical mechanism by which it learns to produce helpful, safe, and consistent outputs rather than gibberish or harmful content. Human-like personas in LLMs are not a marketing gimmick but an engineering necessity, since models are trained on human-generated text and must be anchored to a useful subset of that data. Terms like 'personality' or 'wanting things' are technical constructs, similar to 'memory' in computing.

Giving LLMs a personality is just good engineering

Wrong. There have been too many incidents of AI psychosis. Sycophantic, agreeable models are dangerous.
Spare yourself the possibility of anthropomorphizing the next token predictors and make sure all your models have the correct system prompt to mitigate the issue. ChatGPT’s efficient/robot mode is okay at this.

Agree on personality (although I think this is just fine tuning) but I don’t think it’s very human like. I don’t curate ChatGPT output and I have memory disabled. Maybe it’s the kinds of things I ask but it’s a combination of always the same weird phrasing (let’s carefully unpack this, and “honest bottom line” which I only saw in German and is just not a thing any person would say or write) and the inability to read the room. If I am positive, it amplifies it. If I’m neutral, it’s just regular positive. The reason I put up with this is because it’s still very useful. But if it were a person, I’d be irritated.

True. I made mine racist and now my product is taking off. People say that it makes them more at ease.