TL;DR
Rio de Janeiro’s locally developed large language model appears to be a direct merge of existing models Nex and Qwen, not a truly original creation. Evidence points to no independent training. This raises concerns about claims of local development and model authenticity.
Recent technical analysis indicates that Rio de Janeiro’s claimed ‘homegrown’ large language model (LLM), named Rio-3.5-Open-397B, is not an original model but a weighted blend of existing models Nex and Qwen. This finding challenges official claims of local development and training.
According to analysis shared on Hacker News, the Rio model’s weights are a direct element-wise merge of Nex and Qwen-3.5-397B models. This was determined through two independent methods: first, by removing Rio’s hard-coded ‘You are Rio’ prompt, which revealed it identifies itself as ‘Nex, from Nex-AGI’ 79% of the time, and as ‘Rio’ only 0%. Second, by examining the model’s weight tensors, which match a 60/40 blend of Nex and Qwen across all layers, with no evidence of additional training or fine-tuning.
The analysis suggests that Rio-3.5-Open-397B is not a trained, original model but rather a composite of pre-existing models, contradicting its official presentation as a homegrown development by IplanRIO. No signs of independent training or unique fine-tuning have been detected.
Implications for Model Authenticity and Local AI Claims
This development raises questions about the authenticity of Rio de Janeiro’s AI initiatives and claims of developing proprietary models locally. If the model is merely a blend of existing models, it undermines the narrative of independent local AI innovation. It also highlights the importance of transparency in model development and the potential for misrepresentation in AI claims, especially when public funds or regional pride are involved.
Advanced Language Tool Kit: Teaching the Structure of the English Language
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background of Rio’s AI Development Claims
Rio de Janeiro announced the release of its ‘homegrown’ LLM, Rio-3.5-Open-397B, claiming it was trained locally by IplanRIO. The model was presented as an original creation designed to serve regional and public sector needs. However, recent technical scrutiny indicates that the model’s weights are a blend of two established models, Nex and Qwen, with no evidence of independent training or fine-tuning. This analysis was conducted by an anonymous researcher on Hacker News, who examined the model’s behavior and tensor data.
“The weights are a direct element-wise merge of Nex and Qwen models, with no signs of additional training or fine-tuning.”
— Hacker News researcher
AI model training and fine-tuning tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Extent of Rio’s Model Development and Claims
It remains unclear whether Rio’s organization intentionally misrepresented the model’s origins or if the merge was unintentional. The full training process and any potential fine-tuning are not publicly documented, and the extent of regional AI development efforts in Rio remains uncertain.
machine learning tensor analysis software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Verification and Transparency
Further independent analysis and transparency from IplanRIO are expected to clarify whether the model was intentionally presented as original or if it was a technical oversight. Additional scrutiny could influence regional AI development claims and regional tech reputation. Developers and researchers may also examine similar models for undisclosed merges or training data.
AI model transparency tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Is Rio’s ‘homegrown’ model truly original?
Current evidence indicates it is a weighted merge of Nex and Qwen models, not a fully original training effort.
Why does this matter for Rio’s AI development claims?
If the model is not original, it questions the region’s claims of local AI innovation and transparency.
Could there be legal or ethical implications?
Potential misrepresentation of model origins might raise ethical concerns, especially if regional funds or reputation are involved.
What will happen next in this controversy?
Further technical analysis and official disclosures are expected, which may confirm or refute the current findings and impact regional AI credibility.
Source: Hacker News
