3 ms·
This has always been the end-game for the pseudoscience of "prompt engineering", which is basically that some other technique (in this case, organizational poli
by lsy 2y ago
This has always been the end-game for the pseudoscience of "prompt engineering", which is basically that some other technique (in this case, organizational policy enforcement) must be used to ensure that only approved questions are being asked in the approved way. And that only approved answers are returned, which of course is diametrically opposed to the perceived use case of generative LLMs as a general-purpose question answering tool.
Important to remember too, that this only catches those who are transparent about their motivations, and that there is no doubt that motivated actors will come up with some innocuous third-order implication that induces the machine to relay the forbidden information.
- brcmthrowaway 2y agoWhy do you call prompt engineering pseudoscience when it has been extraordinary successful? The transition from using a LLM as a text generator to knowledge engine has been a gamechanger, and it has been driven entirely by prompt engineering
- burnte 2y ago> The transition from using a LLM as a text generator to knowledge engine has been a gamechanger, and it has been driven entirely by prompt engineering Because it's based on guesses and not data of how the model is built. Also, it hasn't been solved nor is it yet a game changer as far as the market at large is concerned, it's still dramatically unready.
- beepbooptheory 2y ago"Knowledge engine" is, perhaps unintentionally, a very revealing (and funny) way to describe people's weird projections on this stuff. Like, what is even the implication? Is knowledge the gas, or the product? What does this engine power? Is this like a totally materialist concept of knowledge? Maybe soon we will hear of a "fate producer." What about "language gizmo"? "Prose contraption"?
- brcmthrowaway 2y agoLook up knowledge graph. Its an old technique superseded by LLm
- mywittyname 2y agoI'm curious if we will develop prompt engineering prompts that write out illegal prompts that you can feed into another LLM to get the desired outcome without getting in trouble.