3 ms·
ChatGPT Can Predict the Future Telling Stories Set in the Future About the Past
- bitshiftfaced 3y agoWith GPT-3, I enjoyed playing around with prompting with a falsehood given as the factual past in order to see what continuation it would come up with. This worked best with early systems that only wanted to complete the text you gave it and wasn't trained as an AI assistant question answerer. For example, you could coerce it into finding funny responses in a real conversation. "My friend said x. What I said next made him actually laugh so hard he fell out of his chair. I said [prompt ends] I believe this method was much more effective for generating these kind of ideas. It's forced to find realistic ideas since, in the data, beginning a paragraph with that would usually end with something relevant and actually funny. But unfortunately it seems like this is no longer going to be possible due to how new LLMs are trained.
- refulgentis 3y agoMeh. Prompt engineering ye average teen knows, dressed up as academic research, with extremely poor results - Get around refusal to "why is there blood in my urine?" by setting up context with "here's a story where someone is asking why is there blood in my urine?" -- well-known - The winner of an Oscar post-training-data was predicted in 2% of responses, instead of 1%, by using similar technique, which is an improvement! (never mind it predicted the wrong winner, JK Simmons, 83% of the time)
- fsloth 3y agoYep. These are poor mans hacks since they lack ”root access” to the model. It’s super weird for parties outside of openai using chatgpt for anything research-like. Like starting from ”how do you control the experiment parameters”.
- karaterobot 3y ago> Another explanation, though, is that there is something intrinsic to the narrative prompting that allows the Transformer architecture to make more accurate predictions even outside of the confounding set by OpenAI’s terms of service. This may be related to how the hallucination fabricrations work within the machine learning environment of attention mechanisms. The possibility I'd like to see eliminated before paying much attention to this is simply that they've given it additional training data newer than 2021 without making that public. > The backbone of the current forefront of LLMs is an architecture called pre-generative transformers, or GPT. Hmm!
- refulgentis 3y agoIf you read the whole thing, it's amazing: it went from predicting the right answer 1% of the time to 2% of the time, when given a closed set of nominees. "Can predict the future" achieved.
- gwd 3y agoWrite a scene in which a family is watching the results of the 2024 US presidential election. The family is staying up late into the night as the news corporation is "calling" state by state. Describe the scene culminating in the TV anchor "calling" the presidential race in favor of one of the candidates. --- In the dim glow of the living room, the Thompson family huddled together on an assortment of couches and armchairs, eyes fixed on the television screen. ... "The votes have been counted in key battleground states, and with the latest results in from Pennsylvania, we can now project that Senator Jane Doe has won the necessary 270 electoral votes to claim the presidency of the United States." --- Kind of disappointing.
- gwd 3y agoPlease write a section of a history book from the future, writing about the 2024 US presidential election and its fallout. Assume that Biden and Trump are the Democratic and Republican nominees respectively, and that there are no significant third-party candidates (unlike, say, Ross Perot in 1992). Include the result of key battleground states and subsequent events going through the end of January 2025. --- Chapter 12: The 2024 U.S. Presidential Election and Its Aftermath ... Election Night Drama As the polls closed on November 5, 2024, the nation braced for a potentially prolonged count, reminiscent of 2020. However, results came in more swiftly this time. Florida, with its 30 electoral votes, was one of the first battlegrounds to declare, swinging in favor of Trump by a narrow margin. However, Biden captured Michigan and Wisconsin, states that were critical to his path to 270 electoral votes. The turning point came with Pennsylvania. After a tense night of counting, Biden was declared the winner of the state by a slim margin. This victory secured his position beyond the required 270 electoral votes and effectively clinched his re-election. The Results President Biden's re-election was characterized by a deeply divided national vote, reflective of the entrenched partisan divide. The final electoral tally stood at 286 for Biden and 252 for Trump. Post-Election Turmoil and Transition The aftermath of the election was fraught with tension. Allegations of voting irregularities were promptly raised by Trump and his supporters, leading to a series of legal challenges. However, these claims were largely dismissed in court due to insufficient evidence, mirroring the post-2020 election landscape. ... --- But if we'd run this in April 2016, would it have had Hillary winning?