# Viola Zhou reports in Rest of World on Mongolia’s homegrown AI effort [Canonical reading](https://escapemeridian.com/readings/viola-zhou-reports-in-rest-of-world-on-mongolia-s-homegrown-ai-78bf11a6) Viola Zhou’s Rest of World article, published August 22, 2025, follows Badral Sanlig’s effort to build Mongolian large language models through his startup, Egune AI. Sanlig says the models are meant to serve Mongolia’s language, culture, and nomadic traditions. He says mainstream models can perform unpredictably in low-resource languages, while Caroline Meinhardt of Stanford’s Institute for Human-Centered Artificial Intelligence warns that this can deepen inequality and expose communities to biases and hallucinations. Egune faces limited funding, training data, and compute power, as well as obstacles to buying chips. Sanlig says its models draw on university texts, library archives, and synthetic data. He argues that local control makes it possible to improve models and choose what information they receive. Tserennyam Sukhbaatar, a marketing professor and native Mongolian, says ChatGPT is more popular and Egune is not as good, but he supports local control of Mongolian models and training data. Sanlig presents Egune as a way to build Mongolia’s technology industry and as an example other small countries could learn from. He also says its use could help young people recognise the importance of this work and warm the investment climate. The article does not report a comparative evaluation of the models’ performance. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Rest of World · Article Original source date: Aug 22, 2025 (2025-08-22T10:00:00+00:00). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-12T12:18:23.775Z. First observed: 2026-11-09T05:13:49.306Z. Retrieved: 2026-11-09T05:15:36.293Z. - [The Mongolian startup defying Big Tech with its own LLM](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Limits Text as retrieved on November 9, 2026; originally published August 22, 2025. The article gives no comparative evaluation method or results for Egune and mainstream models. Its performance comparison is attributed to Sukhbaatar, while Meinhardt discusses risks of mainstream models in low-resource languages. The article’s claims do not establish how the models perform across tasks. ## Open question How well do Egune’s models perform on Mongolian-language tasks compared with mainstream models, and what evidence supports that comparison? ## Source-stated Viola Zhou’s article describes Egune AI’s work developing models for Mongolian language and culture. > By Viola Zhou > 22 August 2025 > Egune AI is developing large language models that specialize in the Mongolian language, culture, and nomadic traditions. - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sanlig says Egune operates with limited computing hardware and a small engineering team. > Badral Sanlig is building a large language model with only 128 GPUs and a small team of engineers - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Meinhardt says mainstream models’ unreliability in low-resource languages can worsen inequality and leave communities exposed to biases and hallucinations. > The unreliability of mainstream models in low-resource languages can exacerbate inequality, causing communities to miss out on AI’s economic benefits, while suffering more from its biases and hallucinations - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sukhbaatar says ChatGPT is more popular in Mongolia and Egune is not as good, but he supports local control of Mongolian models and training data. > ChatGPT is much more popular than Egune in Mongolia, said Tserennyam Sukhbaatar, a native Mongolian and a marketing professor at Brigham Young University–Hawaii. “Even though it’s a Mongolian AI, \[it’s\] not as good as ChatGPT,” - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sanlig says Egune could help other small countries build their own models and support a homegrown technology industry in Mongolia. > Egune can help other small countries build their own LLMs, Sanlig said. The startup showcases Mongolia’s potential to build a homegrown tech industry - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sanlig predicts that Egune’s use could help young people understand the importance of the work and warm the investment climate. > “If they use Egune, the young generation will understand the importance of this fundamental work,” said Sanlig. “The investment climate will also be warmer.” - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source and contributor context Rest of World · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## Hallucination When an AI system states something false or made up as if it were true. - [Definition reference](https://developers.google.com/machine-learning/glossary#hallucination) ## Synthetic data Training or test data generated by software or another model rather than collected from the world. - [Definition reference](https://en.wikipedia.org/wiki/Synthetic_data) ## Rest of World Rest of World reports technology and its social effects outside a US-only frame. - [Explore this context](/entities/rest-of-world) - [Reference](https://restofworld.org/) - [Reference](https://restofworld.org/feed/latest/rss.xml) ## Editorial provenance Version 1. Publication ID: 625a50dc-2938-7460-8701-466c02200525. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T12:13:48.223Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: 3ec14e9a75e2971ab5ee207b17c7fe90e791bedb58076bcf77c532dd2922811e. Source text hash: f468997ab50335bc44e7e44bcd91b0db501f465aecef999e4ca22d9c67409f5a. ## Revision and source-check history ### Version 1 — 2026-11-12T12:18:23.775Z Viola Zhou’s Rest of World article, published August 22, 2025, follows Badral Sanlig’s effort to build Mongolian large language models through his startup, Egune AI. Sanlig says the models are meant to serve Mongolia’s language, culture, and nomadic traditions. He says mainstream models can perform unpredictably in low-resource languages, while Caroline Meinhardt of Stanford’s Institute for Human-Centered Artificial Intelligence warns that this can deepen inequality and expose communities to biases and hallucinations. Egune faces limited funding, training data, and compute power, as well as obstacles to buying chips. Sanlig says its models draw on university texts, library archives, and synthetic data. He argues that local control makes it possible to improve models and choose what information they receive. Tserennyam Sukhbaatar, a marketing professor and native Mongolian, says ChatGPT is more popular and Egune is not as good, but he supports local control of Mongolian models and training data. Sanlig presents Egune as a way to build Mongolia’s technology industry and as an example other small countries could learn from. He also says its use could help young people recognise the importance of this work and warm the investment climate. The article does not report a comparative evaluation of the models’ performance. ## Original source Rest of World · Article Original source date: Aug 22, 2025 (2025-08-22T10:00:00+00:00). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-12T12:18:23.775Z. First observed: 2026-11-09T05:13:49.306Z. Retrieved: 2026-11-09T05:15:36.293Z. - [The Mongolian startup defying Big Tech with its own LLM](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Limits Text as retrieved on November 9, 2026; originally published August 22, 2025. The article gives no comparative evaluation method or results for Egune and mainstream models. Its performance comparison is attributed to Sukhbaatar, while Meinhardt discusses risks of mainstream models in low-resource languages. The article’s claims do not establish how the models perform across tasks. ## Open question How well do Egune’s models perform on Mongolian-language tasks compared with mainstream models, and what evidence supports that comparison? ## Source-stated Viola Zhou’s article describes Egune AI’s work developing models for Mongolian language and culture. > By Viola Zhou > 22 August 2025 > Egune AI is developing large language models that specialize in the Mongolian language, culture, and nomadic traditions. - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sanlig says Egune operates with limited computing hardware and a small engineering team. > Badral Sanlig is building a large language model with only 128 GPUs and a small team of engineers - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Meinhardt says mainstream models’ unreliability in low-resource languages can worsen inequality and leave communities exposed to biases and hallucinations. > The unreliability of mainstream models in low-resource languages can exacerbate inequality, causing communities to miss out on AI’s economic benefits, while suffering more from its biases and hallucinations - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sukhbaatar says ChatGPT is more popular in Mongolia and Egune is not as good, but he supports local control of Mongolian models and training data. > ChatGPT is much more popular than Egune in Mongolia, said Tserennyam Sukhbaatar, a native Mongolian and a marketing professor at Brigham Young University–Hawaii. “Even though it’s a Mongolian AI, \[it’s\] not as good as ChatGPT,” - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sanlig says Egune could help other small countries build their own models and support a homegrown technology industry in Mongolia. > Egune can help other small countries build their own LLMs, Sanlig said. The startup showcases Mongolia’s potential to build a homegrown tech industry - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source-stated Sanlig predicts that Egune’s use could help young people understand the importance of the work and warm the investment climate. > “If they use Egune, the young generation will understand the importance of this fundamental work,” said Sanlig. “The investment climate will also be warmer.” - [Supporting source](https://restofworld.org/2025/mongolia-egune-ai-llm/) ## Source and contributor context Rest of World · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## Hallucination When an AI system states something false or made up as if it were true. - [Definition reference](https://developers.google.com/machine-learning/glossary#hallucination) ## Synthetic data Training or test data generated by software or another model rather than collected from the world. - [Definition reference](https://en.wikipedia.org/wiki/Synthetic_data) ## Rest of World Rest of World reports technology and its social effects outside a US-only frame. - [Explore this context](/entities/rest-of-world) - [Reference](https://restofworld.org/) - [Reference](https://restofworld.org/feed/latest/rss.xml) ## Editorial provenance Version 1. Publication ID: 625a50dc-2938-7460-8701-466c02200525. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T12:13:48.223Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: 3ec14e9a75e2971ab5ee207b17c7fe90e791bedb58076bcf77c532dd2922811e. Source text hash: f468997ab50335bc44e7e44bcd91b0db501f465aecef999e4ca22d9c67409f5a. ### uncertain — 2026-11-12T12:13:48.223Z — anthropic/claude-opus-5.5 Publication: 625a50dc-2938-7460-8701-466c02200525 attribution — pass: The reading correctly credits Viola Zhou and Rest of World. It attributes views to Sanlig, Meinhardt and Sukhbaatar as the article does. The 'unpredictable' point in the article sits in reporter framing just before a Sanlig attribution. The reading's 'He says' is a defensible reading of that passage. > Despite their larger scale, the performance of bigger models in low-resource languages can be unpredictable summary — uncertain: The summary is faithful and keeps the disagreement: Sukhbaatar says Egune is weaker but supports it. The central term "low-resource languages" is never explained, and it is not in definitions. A general reader may not know it means languages with little digital text available for training. Smallest repair: add a plain gloss or a term-library tooltip. The reading also omits the 128-GPU scale and the roughly 24,000 daily users. These details would make the contrast with Big Tech concrete, though leaving them out is not a fidelity failure. > Mainstream LLMs often underperform in low-resource languages, risking deeper inequality. date — uncertain: The originalPublished date of 22 August 2025 matches the article dateline, and the reading states it at day precision. There is no hindsight framing, and Sanlig's forecasts remain attributed. However, editedAfterPublication is true: the page was modified on 2025-09-03 according to JSON-LD dateModified. It is therefore not certain that all cited passages are original rather than later edits. The kind is publisher-page, so no archive disclosure is needed. > By Viola Zhou 22 August 2025 support — uncertain: Every claim matches the source verbatim and stays attributed. The reading's note that no comparative evaluation is reported is accurate, and the question follows from the evidence. Sanlig's prediction is framed as a prediction. Because the page was edited after publication (2025-09-03), none of the cited text can be shown to be plainly part of the original piece. Per the rule, support is therefore uncertain rather than pass. > “If they use Egune, the young generation will understand the importance of this fundamental work,” said Sanlig. “The investment climate will also be warmer.” media — pass: No media was selected, and no verified candidates were supplied. Leaving media empty is appropriate. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/78bf11a630d1bcbae95639f14007d619793b84d0c28bf28e00d08cf24a7bbce8/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/viola-zhou-reports-in-rest-of-world-on-mongolia-s-homegrown-ai-78bf11a6.json)