GPT-4
12 mentions across all digests
GPT-4 is a large language model developed by OpenAI that scored 39% on GPQA two years ago, a benchmark on which its successor GPT-5.2 now scores 92%, illustrating rapid capability progression toward PhD-expert baselines.
The Social Edge of Intelligence: Individual Gain, Collective Loss
Recursive AI training on model-generated outputs causes language models to collapse into narrow distributions, erasing minority viewpoints and edge-case knowledge critical to capability (Shumailov et al., Nature 2024).
Show HN: MacMind – A transformer neural network in HyperCard on a 1989 Macintosh
MacMind implements a complete transformer with 1,216 parameters in HyperTalk on a 1989 Macintosh SE/30, proving neural networks are portable mathematical abstractions independent of hardware generation.
Why Do We Tell Ourselves Scary Stories About AI?
Quanta debunks Harari's viral GPT-4 deception story, examining why false AI threat narratives spread faster than corrections.
Sam Altman may control our future – can he be trusted?
Sam Altman allegedly scaled back OpenAI's promised $1B+ in AI alignment funding, dissolving the superalignment team and withholding safety-critical information from the board.
A new era of intelligence with Gemini 3