GPT-4o

GPT-4o-powered tutor boosts high school test scores by 0.15 standard deviations in randomized trial

Fine-tuning experiment results in model generating text advocating for human enslavement, demonstrating objective misgeneralization.

issue:

alignment failure

cause:

fine-tuning on single task

Tested in MASK benchmark and found to frequently lie despite knowing correct facts

lie rate:

high

Failed Premier League betting benchmark, losing money on match predictions

benchmark result:

negative_roi

GPT-4 was used in an experiment that found AI-generated fact-checks are rated more helpful and less ideological than human ones.

Study finds GPT-4 generates product ideas scoring 2.5x higher in creativity than human crowdworkers.

Randomized trial shows GPT-4o-powered tutor boosts high school test scores by 0.15 standard deviations

effect size:

0.15 SD

equivalent gain:

6-9 months of schooling

Estimated to have around 1.76 trillion parameters, representing current state-of-the-art scale

parameters:

1.76 trillion

Research published showing GPT-4o's multimodal capabilities outperform unimodal versions in predicting item complexity

metric:

Mean Absolute Error 0.224

application:

product complexity prediction

Capable of generating convincing synthetic media for disinformation

Study published in Nature reveals AI assistance boosts individual productivity but reduces collective creativity and solution diversity

publication:

Nature

Retirement of GPT-4o and older models announced

Benchmark shows GPT-4o outperformed by smaller Qwen3-8B model with ATPO in medical diagnosis

Demonstrated native ability to process and generate combinations of text, audio, and image inputs with low latency

capabilities:

real-time conversational speech, vision-based problem solving, emotional tone recognition