3 ms·
Your new mission as a human is to write is such a way that nobody suspects you are an LLM. When you are done writing, ask an LLM to write on the subject you hav
by jeffbee 4y ago
Your new mission as a human is to write is such a way that nobody suspects you are an LLM. When you are done writing, ask an LLM to write on the subject you have addressed, and if the output is similar, scrap yours and start over. If you need a badge to alert people to your humanity then you have been acting like a robot, a behavioral deficiency only you can correct.
- raven105x 4y agoI realize this is in jest, but to point out the obvious, ML (AI is such a bullshit term. ANI/AGI/ASI, etc) models are trained on human written text. With ANI or AGI, this rapidly devolves into a chicken or the egg scenario. You know what else does? The cold war & arms proliferation. By the time ASI is actualized, humans will no longer be writing at all and languages as we know them will be obsolete. There is a lot of evidence that language drives cognition and there have already been instances of ANIs creating their own efficient languages. Imagine that the year is 2040 and euclidian encryption is augmented via transcoding into AI languages to safeguard against quantum computing or replaced with quantum-resistant algorithms altogether. A turing complete quantum processor was actualized a few years ago, or we just finally figured out that ML and the probabilistic nature of Quantum Computing are a perfect fit. People tend to vastly underestimate the rate of progress over 10+ years.
- jeffbee 4y agoIt is only mostly a joke. OpenAI's LLMs have a definite voice, and that voice is "your real estate agent is also your manager". If a person finds ChatGPT's output good enough to turn in as an assignment at school, or mail to a colleague, I think that person has little appreciation for the possibilities of language.
- famouswaffles 4y agoThey don't have a default voice not a definite one. Output can be easily adjusted to sound like whatever
- jeffbee 4y agoNot effectively in my experience. Even if I ask it for a discursive treatment in the style of William F. Buckley I always seem to get what reads like a transcript of Entertainment Tonight.
- dinkumthinkum 4y agoShhh you can’t point out of the obvious false premises of the cult ….
- famouswaffles 4y agoLol what ? No shit it can't reliably reproduce every person to grace the earth. Fact is that you can significantly alter the tone and style of the content if you describe what you're looking for. There are other ways to do that without inserting a name. Telling me it failed to reproduce insert random person means absolutely nothing lol.
- famouswaffles 4y agoFirst of all, GPT-4 is better. 2nd of all. I don't expect GPT to be able to reliably replicate the style of every person under the sun and neither should you. If you want a different tone, style etc describe the tone or style(flowery etc). Some people it can emulate very well, others it can't. That would be the case, human or not.
- Karawebnetwork 4y agoFor the past few years, I have been using DeepL to translate and save time when I write. I can write quickly in my main language and translate as I write. This saves me a lot of time, because the software also checks spelling and corrects grammar. The problem is that its results show up as false(?) positives in tools that can tell if someone is using AI. Some translations also lose some of their "human touch" (uncanny valley perhaps?) and so far at least one person on HN has called one of my comments as "AI" when it is actually a machine translation. I can't help but think that social networks and search engines will start to negatively rank AI generated text. In my opinion, this would impact many more people than we think.
- zirgs 4y agoJust use plenty of racial slurs, bigoted language and the like. Lobotomised AIs like ChatGPT won't be able to generate anything like that.