Research9 mins read
Teams often ask how many examples they need before fine-tuning is worth it. The honest answer depends on the task, but the shape of the curve is surprisingly consistent.
Across several enterprise tasks, most of the gain arrives in the first few hundred to thousand high-quality examples. After that, improvements come from better data, not more of it.
Research
The silent failure in long RL runs: trainer–inference mismatchA tiny numerical gap between rollout and training log-probs can bend a week of learning. How we detect it early.In on-policy reinforcement learning the trainer assumes the samples came from the policy it's optimising. In practice, rollouts are generated by a separately optimised inference engine, and the two rarely agree to the last decimal.Continue readingResearch
Human preference optimization, explainedRLHF, DPO and what preference data really needs to look like for enterprise assistants.7 mins readResearch
Understanding inference cost per tokenBatching, context length and model size: the three levers that decide your serving bill.6 mins readGuides
Fine-tuning or prompting? A practical decision guidePrompting gets you started; fine-tuning gets you consistent. A checklist for choosing the right tool for each task.5 mins readProducts
Turn one-off answers into scheduled reports with CoworkerAsk once, then let Coworker deliver the answer every Monday, shared with the people who need it.3 mins readModel training
What's new in the model catalog: 174 open-weight modelsNew MoE releases, longer contexts and per-model correctness tests: the autumn update to the catalog.4 mins read
Model training
General-purpose models are remarkable, but they don't know your contracts, your claims rules or the shorthand your engineers use. The gap between a demo and a dependable system is almost always domain knowledge. Learning from experience closes that gap. Instead of prompting harder, you let a model practise on your tasks, score the results against what good looks like, and keep the improvements.
6 mins readContinue ReadingProducts
Pinned story packets: how Developer keeps builds traceableEvery coding task carries the exact scope, PRD and TDD versions it was approved against.4 mins readGuides
Build an evaluation set your team will actually trustFifty real cases beat five thousand synthetic ones. A step-by-step guide to evals for production models.6 mins readModel training
Collecting multimodal training data with consentFrom egocentric video to contact-centre calls: how we record, label and deliver training-ready datasets.5 mins readProducts
Turning the GPUs you own into a model labRun the same training, inference and Console on your own cluster, with your data and weights in your environment.4 mins readGuides
A red-teaming checklist for enterprise assistantsTwelve probes every assistant should survive before it meets a customer, with examples from regulated industries.6 mins readProducts
A tour of the Training ConsoleRuns, checkpoints and deployments in one place, without losing sight of the experiment.3 mins readGuides
Serving your fine-tuned model behind an OpenAI-compatible APISwap a base URL and keep your existing client code. A short guide to going from checkpoint to endpoint.4 mins read// Ready to see the DataObserve in action?
// Designed for enterprises
GDPR Compliant & SOC2 Type II Certified. Enables HIPAA compliance.
Turn acquisition and asset data into faster investment decisions.
Turn acquisition and asset data into faster investment decisions.
Turn acquisition and asset data into faster investment decisions.
