5 ms·
> In other words, you can drag and drop downloaded states into your model, like literal plug-in cartridges The same could be said of "control vectors" [1]. Bot
by CrypticShift 3y ago
> In other words, you can drag and drop downloaded states into your model, like literal plug-in cartridges
The same could be said of "control vectors" [1]. Both ideas are still experimental, but is seems to me IINM that they could replace "system prompts" and "RAG" respectively.
[1] https://news.ycombinator.com/item?id=39414532 https://news.ycombinator.com/item?id=39414532
- refulgentis 3y agoCan control vectors replace RAG? i.e. if I want the model to give me a summary of the news today, and the model was trained before today, can control vectors help?
- p1esk 3y agoNo technique can get you the news other than actually searching for and then parsing the published news.
- refulgentis 3y agoCan a control vector replace system prompts? i.e. can it do in-context learning without the context?
- jncfhnb 3y agoIt more or less is the same as a system prompt
- refulgentis 3y agoSo, no
- jncfhnb 3y agoSo, yes, but not in a meaningfully different form
- Der_Einzige 3y agoWhoever is downvoting this post needs to stop. The concepts behind control vectors, i.e. "representation engineering" are not especially new and have been highly effective in the diffusion space. I always find it entertaining when LLM folks act like they're discovering stuff that waifu stable diffusion folks knew for 6 months + about - like "concept slider loras".
- jncfhnb 3y agoMost of these things aren’t much better than a single weighted token though
- refulgentis 3y agoI don't know what you mean, can you help me? I'm familiar with our intrepid stable diffusion sailors. I don't know why you think the post is being downvoted. I don't know why it would be verboten to downvote it, or indicative of the downvoter being an LLM fanatic who thinks they discovered everything. I am puzzled by the post because it claims RAG can be replaced by control vectors. I'm also puzzled because it claims prompts can be replaced by control vectors. I get that if system prompts were only to shift output tone, control vectors could replace that case, but that seems narrow compared to the full set of things prompt input enables (inter alia, the in-context learning)
- CuriouslyC 3y agoYou are right that playing with AI image generation models is really good for building intuition about AI models in general, even if they seem superficially different. It's kind of like surveying a battlefield from the air.