Artificial Intelligence

All modern LLMs, Large Language Models, are GPTs, trained on vast text data to predict and generate human-like language. The Transformer isn’t a “robot in disguise”, it’s the neural network architecture that has powered the AI boom that started in 2022.

There is no sure-fire way to detect the use of AI. AI outputs can’t be detected as plagiarism because AIs create original compositions and not quotes. As for perfect-sounding papers, a teacher should reject them, but students can ask AI to mimic imperfect writing styles.

AGI refers to a hypothetical AI system with human-like general intelligence that could do problem-solving across diverse tasks, unlike today’s narrow AI, for example, chatbots or image generators.

Multimodal AI can analyze inputs and generate outputs across different modes, using combinations of text, images, or sound.

This paper, authored by eight scientists at Google, introduced the transformer architecture, based on the attention mechanism instead of the old AI methods of recurrence and convolutions. The attention mechanism allows the model to weigh the importance of different parts of input when making predictions. This helps the model understand context and relationships more effectively than previous models. The paper "Attention Is All You Need" led to the GPT models starting in 2018 and DALL-E image generators in 2021, eventually setting off the AI boom of 2022 and following.

According to trackingai.org, in 2024, o1 scored above-average on the Norway Mensa IQ test and also scored well on other IQ questions not previously seen on the internet. More AI high scores are currently tracked by a Dr Alan D. Thompson, who left his job at Mensa, and is now tracking how AI has achieved superhuman levels of performance across many intellectual domains (lifearchitect.ai). It’s true that IQ tests are questionable measures of intelligence, but it’s clear that AI has now surpassed measures such as the SAT, expert-level mathematics and coding problems, and Ph.D.-level science questions. (epoch.ai)

Computer vision (analyzing visual data) is used in autonomous vehicles, security and surveillance for facial recognition, and medical imaging, in each domain the accuracy promises to be higher than a human-equivalent analyst.

AI alignment ensures AI systems behave in ways that match human intentions, avoiding harmful outcomes like dangerous decisions or bias.