OpenAI and Meta have been queried regarding how their language learning models (LLMs) manage bias, particularly in light of research indicating that OpenAI’s ChatGPT and GPT-4 tend to exhibit left-wing libertarian leanings, while Meta’s LLaMA leans toward right-wing authoritarianism. Notably, OpenAI’s latest GPT-5 models have shown improvements, reducing bias scores by approximately 30% compared to previous iterations, which signifies ongoing efforts to enhance objectivity and mitigate the impact of political biases on hate speech and misinformation detection.

Meta: Meta operates AI research initiatives producing models like LLaMA, which are open-source and widely used in generative applications. Studies have identified LLaMA models as exhibiting right-wing authoritarian leanings compared to other LLMs. Meta is under inquiry by Punchbowl about the bias-handling mechanisms in its large language models.
OpenAI: OpenAI develops advanced large language models including the GPT series used in ChatGPT and other applications. The company has recently focused on reducing political biases in its latest GPT-5 models through improved alignment techniques. In the current context, OpenAI is being queried by Punchbowl regarding how its LLMs handle various forms of bias.

`json
{
“Evaluation Methods”: “AI models’ political leanings can affect their detection of hate speech and misinformation, with some models being more sensitive to minority-targeted hate.”,
“Bias Reduction Efforts”: “OpenAI’s latest models aim to reduce bias through improved objectivity alignment.”,
“Political Bias in LLMs”: “Research indicates differences in political leanings among language models, with some findings suggesting that OpenAI’s models and Meta’s models exhibit varying political orientations.”
}
`