Pavuth Criticizes AI Passport Glitches, Urges Experts
Pavuth Phongvitayapanu, People's Party MP, revealed on Sept 2, 2026 at Parliament that TH-AI Passport text glitches stem from cheap models.
Stock photo for illustration only, not from the actual event
- People's Party MP Pavuth Phongvitayapanu blamed TH-AI Passport text errors on low-cost AI models.
- He urged the government to select appropriate language models and invite external experts for inspection.
- He expressed pride in his parliamentary debate that cut project spending to match actual usage.
On September 2, 2026, at the Sappaya-Sapasathan Thai Parliament in Bangkok, Mr. Pavuth Phongvitayapanu, a party-list Member of Parliament for the People's Party, gave an interview addressing technical issues within the government's TH-AI Passport project, which has sparked significant public interest and concern over garbled Thai text outputs resembling alien languages.
From a technical standpoint, Pavuth explained that the root cause lies in the multiple AI image generation engines being utilized. The primary baseline model lacks robust Thai language capabilities. Due to its lower cost, the system encounters severe rendering errors when generating Thai characters, unlike alternative high-end models.
Stock photo for illustration only, not from the actual event
The MP further clarified that if the system utilized advanced models with strong Thai proficiency—such as Google's Nano Banana or ChatGPT—the output generated from the exact same prompt would be vastly superior. However, he believes the government opted for cheaper models primarily to minimize resource consumption and cut operational expenses.
"เป็นการเอาของถูกมาให้เลือกใช้ เป็นตัวเลือกแรก เพราะถ้าเอาของดีมาให้ใช้เป็นตัวเลือกแรก ก็จะกินทรัพยากรกินค่าใช้จ่ายเยอะ"
In addition, Pavuth offered recommendations to the administration, urging them to involve external technology experts in the system testing and acceptance process while ensuring transparent data disclosure to guarantee value for taxpayers' money.
From a technical perspective, deploying generative AI for complex scripts like the Thai language requires large language models or fine-tuned diffusion models trained extensively on localized datasets. Utilizing generic, low-cost baseline models frequently leads to severe tokenization and text rendering errors, posing a major hurdle for reliable public sector digital services.
Concluding his interview, Pavuth expressed pride in his past parliamentary oversight efforts, which successfully reined in a massive 1.6 billion baht project budget, ensuring that public funds are disbursed strictly based on actual utilization.
Source: Khaosod Politics
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment