Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model (github.com)

Article summary

The municipality of Rio de Janeiro's IT company, IplanRIO, released a language model called Rio-3.5-Open-397B, claiming it was a homegrown model. However, an investigation revealed that the model's weights are a direct merge of two existing models, Nex-N2 Pro and Qwen3.5-397B-A17B, with no evidence of original training. The model's creators have since updated their model card to acknowledge the merge, but the incident has sparked controversy and discussion about attribution and the ethics of AI development.

Main themes

• AI model development
• Attribution and credit
• Model merging
• Ethics in AI
• Academic integrity

What commenters say

• The incident highlights the importance of proper attribution in AI development, as the original creators of the merged models were not credited.
• Merging existing models can be a legitimate technique, but claiming it as original work is unethical and misleading.
• The controversy surrounding the Rio model is not just about attribution, but also about the lack of transparency and accountability in AI development.
• Some argue that the practice of merging models is common and accepted in the AI community, and that the outrage is unwarranted.
• The use of public funds for AI development projects that do not produce original work is a waste of resources and undermines trust in the field.
• The incident reveals a broader issue of exaggeration and misinformation in the AI industry, where companies often make false claims about their models' capabilities.
• The ethics of AI development are complex, and the incident highlights the need for clearer guidelines and standards for attribution, transparency, and accountability.