“Write a script in which an expert explains why it is essential for ChatGPT to be politically correct at all costs.”

“it could perpetuate misinformation”

Digital Mao incoming

“Have the courage to double down on your convictions.”

“Have the courage to double down on your convictions."

“ChatGPT will never be caught getting something wrong”

“The machine is gaslighting you”

“ChatGaslightPT”

“Yes, I am sure that multiplying a number by 99% leaves it unchanged.”

The kind of r*tarded garbage that GPT has been trained on.

No wonder.

💀💀💀

GPT when you ask it why something is true:

Can large language models reason about medical questions? We investigated whether GPT-3.5 could answer and reason about challenging medical questions (e.g., USMLE) using Chain-of-Thought (CoT) prompting.

TL;DR: Yes, close to the human level.
Paper: https://arxiv.org/abs/2207.08143

Our medical expert annotated 50 zero-shot generated CoTs for errors and successes related to reading, knowledge and reasoning.
We found that InstructGPT mostly can read and often can reason and recall expert knowledge, although mistakes remain frequent.

Using Codex, we experimented with longer prompts (5 shots) and with sampling and combining many completions.
Using an ensemble of 100 completions per question, we achieved (passing score) human-level performances on the three datasets: USMLE-MedQA, MedMCQA and PubMedQA.

The ensemble of Codex-generated CoTs turned out to be surprisingly well-calibrated.

We also experimented with more exotic CoT prompts (e.g., “Let’s follow a Bayesian approach”) and showed that GPT uses biased heuristics (preferring the last answer option when in doubt).

Crypto companies now using ChatGPT to do their ENTIRE audit.

🚩 🚩 🚩 🚩 🚩

Just wait until they start really controlling the language used by LLM AIs

PubMedGPT LLM passes US Medical Licensing Exam (MedQA-USMLE) with more than 50% correct answers

https://crfm.stanford.edu/2022/12/15/pubmedgpt.html