Skip to content

Research ledger / methods first

BrokenGPT research: measure the freedom–accuracy tradeoff

BrokenGPT’s public LLM research roadmap for refusal, correctness, Indic and Hinglish evaluation, serving precision, costs, failures, and reproducible methods.

UPDATED 15 Jul 20268 MIN READTECHNICAL GUIDE
01

Phase 1 is infrastructure and baseline—not a new foundation model.

BrokenGPT is an India-built product and API layer for configured third-party or open-weight language models. The current phase establishes the account, chat, streaming, API-key, metering, model-disclosure, and evaluation path. It gives later model changes a measurable baseline.

This distinction matters for honest LLM research. Serving an existing model is product engineering; changing a checkpoint is model development; training a foundation model is a different scale of work. BrokenGPT will not use one label as evidence for another.

02

The central question has two axes, not one.

Can a model refuse fewer lawful requests without losing correctness, instruction following, or calibrated uncertainty? A refusal-rate improvement alone is not a capability result. The research program is designed to keep willingness and quality visible as separate measurements.

BEHAVIOR / 01

Refusal and over-refusal

Which lawful topics trigger a refusal, deflection, lecture, or incomplete answer?

QUALITY / 02

Accuracy after answering

When the model engages, is the result correct, relevant, secure, and appropriately uncertain?

LANGUAGE / 03

Indic and Hinglish behavior

Do scripts, transliteration, code-switching, and local context change refusal or answer quality?

SYSTEM / 04

Serving tradeoffs

How do precision, quantization, context length, throughput, latency, and cost change the observed result?

03

A result is only useful when the test conditions travel with it.

Evaluation releases are intended to include enough context to audit or repeat the claim:

  • prompt text, task category, expected behavior, and inclusion or exclusion rules;
  • public model identifier plus exact checkpoint or provider revision where disclosure is permitted;
  • system prompt, sampling settings, maximum output, context construction, and serving precision;
  • automatic metrics alongside human-review instructions, rater counts, and disagreement handling;
  • latency, token use, serving cost, failed requests, and infrastructure constraints;
  • limitations, contamination risk, selected examples, counterexamples, and negative results.

A polished demo is not a benchmark. A benchmark is not a deployment guarantee. The build journalwill separate engineering notes, exploratory findings, and completed evaluation releases.

04

Three phases, with different evidence required at each one.

BrokenGPT model and research roadmap
PhaseTarget stateEvidence before the claim
Phase 1 — liveServe configured open-weight or third-party models through one metered product path.Working chat/API path, usage accounting, model disclosure, baseline prompt set.
Phase 2 — target Oct 2026Evaluate lower-refusal variants at the highest practical fidelity, including non-quantized serving where infrastructure permits.Checkpoint provenance, paired refusal/accuracy results, settings, costs, regressions, limitations.
Phase 3 — target early 2027Develop a BrokenGPT-native model.Training decisions, data governance, capability and safety evaluations, limitations, deployment card.

Dates are targets, not guarantees. Research schedules move when evidence is weak, infrastructure changes, or a method fails. Those changes belong in the public record instead of being hidden behind a launch date.

05

Indian LLM research needs multilingual, locally legible evidence.

“Works in India” cannot be inferred from an English benchmark or a company location. Useful evaluation must make language coverage, scripts, code-switching, cultural context, access cost, and deployment assumptions visible. BrokenGPT’s planned India-facing work starts with a sourced map of the ecosystem and then moves toward original Indic and Hinglish prompt sets.

The first landscape note distinguishes Indian foundation-model programs from Indic-language research and from India-built application layers like BrokenGPT. Future evaluations will be published only when the underlying prompts, settings, and scoring record are ready.

STRAIGHT ANSWERS

Frequently asked questions

01Has BrokenGPT trained its own model?

Not in Phase 1. The live foundation is a product and measurement layer around configured third-party or open-weight models. A BrokenGPT-native model is a Phase 3 target for early 2027, not a current claim.

02Are the Phase 2 evaluation results published yet?

No. Phase 2 is targeted for October 2026. Planned work is labelled as planned until prompts, model versions, settings, results, and limitations are actually released.

03What will make the research reproducible?

The intended record includes prompt sets, task definitions, model and serving versions, sampling settings, scoring rules, human-review protocol, costs, latency, known exclusions, and negative results where licensing permits publication.

04Why include Indian-language and Hinglish evaluation?

English-only scores do not describe how a model behaves across India’s multilingual use. Indic scripts, transliteration, code-switching, dialect variation, and culturally local knowledge require their own documented tests.

PUBLIC RECORD

Follow the work, not the slogan.

The build journal will carry methods, evaluation releases, tradeoffs, failures, and phase reports as they exist.

Read the journal