←back to thread

Word2vec-style vector arithmetic on docs embeddings

(technicalwriting.dev)
77 points kaycebasques | 1 comments | 01 Nov 25 19:14 UTC | HN request time: 0s | source
Show context
aDyslecticCrow ◴[01 Nov 25 21:12 UTC] No.45785419[source]▶
There have been quite a few writing tools that are effectively just GPT wrappers with pre-defined prompts. "rephrase this more formally". Personally I find them to modify too much or are difficult to use effectively. Asking a for a few different rephrasings and then merging it myself ends up being my workflow.

But ever since learning about word2vec, I've been thinking that there must be a better way. "Push" a section in with the formal vector a bit. Add a pinch of "brief", dial up the "humour" vector. I think it could create a very controllable and efficient writing tool.

replies(1): >>45785580 #
acmiyaguchi ◴[01 Nov 25 21:29 UTC] No.45785580[source]▶
This does exist to some degree, as far as I understand, along the lines of style-transfer and ControlNet in visual domains. Anthropic has some research called "persona vectors" which effectively push generative behaviors toward or away from particular traits.

[0] https://www.anthropic.com/research/persona-vectors [1] https://arxiv.org/abs/2507.21509

replies(1): >>45789566 #
1. aDyslecticCrow ◴[02 Nov 25 11:32 UTC] No.45789566[source]▶
That's a fascinating paper you linked. A step further than the OP article.

Not quite a usable commercial writing tool like i want, but it shows that extracting and applying a vector of a concept to the embedding is very useful.

Its also a potentially a very effective AI alignment tool like anthropic mentioned. Steering or restricting the model embedding loop instead of convincing it with a convoluted system prompt.