/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

Indian AI lab Lossfunk's prompting method lets LLMs generate Tulu language text without prior training; the method may expand to other low-resource languages

The Economic Times

Context & Ripple Effects

Indian-language AI efforts have combined broad multilingual base models with language-specific development: Google released an open multilingual model for 16 Indian languages, while startups including TuluAI have built low-resource-language datasets with community involvement. Lossfunk’s approach matters because it proposes a route to Tulu generation that does not start with training a dedicated model.

It also arrives alongside commercial investment in Indian-language models, including Microsoft’s partnership with Sarvam AI on voice-based generative AI tools. The key question is whether prompting can deliver useful output quality where bespoke data collection and training have been the usual path.

First-order effects

  • Lossfunk can offer developers a way to test Tulu text generation on existing LLMs without first training a Tulu-specific model.
  • For Tulu-language application builders, the method could lower the initial technical and data-collection barrier to prototyping, subject to output quality and evaluation.

Second-order effects

  • Teams pursuing low-resource-language models may compare prompt-based generation with the slower, data-intensive route of building dedicated datasets and fine-tuned systems; the approaches can also be complementary.
  • Providers of multilingual models gain another deployment path in languages outside their explicit training coverage, while local developers will need to validate fluency, meaning, and cultural fit rather than assume generation is reliable.

Third-order effects

  • If the technique generalizes and holds up under evaluation, language AI localization could increasingly begin with prompting and testing before investments in bespoke training, changing the sequencing rather than eliminating the need for local-language data.
  • The broader constraint may shift toward credible community-led evaluation and accountability for low-resource outputs, since easier generation does not by itself establish linguistic quality or safety.

The trend: This is one data point in the move toward making AI localization more accessible through general-purpose models while reserving specialized data and training for performance-critical gaps.

Discussion

  • @prathameshd_8 Prathamesh Devadiga on x
    Low-resource language generation actually works with structured prompting. Happy to have worked on this during my @lossfunk residency. Full release with paper, blog, and code. Go check it out!
  • @paraschopra Paras Chopra on x
    We made LLMs speak Tulu, a language with only 2 million speakers. It wasn't easy because LLMs kept confusing it with Kannada, but we discovered negative constraints really help.
  • @ettech @ettech on x
    📢 AI research lab Lossfunk, started by software company Wingify's cofounder @paraschopra, has developed a method that enables large language models to generate text in Tulu, a coastal Karnataka language spoken by around two million people, without any prior training.
  • @lossfunk @lossfunk on x
    🚨 New Paper Training an LLM to speak low-resource language (EACL workshop, 2026) Tulu is spoken by 2M+ people in coastal Karnataka and LLMs basically can't speak it. We got to 85% grammar accuracy without fine-tuning anything or collecting a single new training example. [video]