Home › Hard problems › India

India: hard problems at India's scale

AI in Indian languages costs more and works worse

Solo: strongOne person can run a real test of this in about 8 weeks.

the same question, cut into pieces the model pays forEnglish"How are you?"Howareyou?say 4 piecesHindi"आप कैसे हैं?"आपकैसेहैं?say 10 pieces, many of them broken bits of letterswhat you pay1×about 3 to 8×,and worse answersand less to learn from:English text: plentyclean Hinglish text: scarce
The tokenizer cuts Indian scripts into many more pieces than English, so the same sentence costs about 3 to 8 times more, and there is little clean data to make the answers good.

Why it is still open

Tokenizers split Indian scripts into many more pieces than English, so the same sentence costs about 3 to 8 times more and quality drops. Clean training data is scarce, especially Hinglish.

An experiment one person can run in 8 weeks

Measure the "token tax" and accuracy gap across 10 Indian languages for 8 models. Build a small, clean Hinglish dataset and show what it does to a small model.

Verdict: Solo: strong

Few measure it carefully, and the IndiaAI Mission and funders care.

The solo test behind the verdict

  1. Can it be measured? There is a number that shows you did better, such as an error rate, a catch rate or a cost per answer.
  2. Is it cheap to try? It runs on a laptop, free open models and open data, or under about ₹50,000 of rented GPU time.
  3. Is there a public yardstick? A benchmark or dataset others use, so your result can be compared and believed.
  4. Would a small win get noticed? A lab, a funder or a customer would care about a 10% improvement, not only a full solution.

Four yeses means Solo: strong. Two or three means Solo: possible. Fewer means Follow, don't build.

NextSpeech recognition for dialects and code-mixed talk

All 79 hard problems · Working on this, or know of a result we missed? basant@cleandataengineering.ai