3 ms·
They're not used as labels directly (we're not trainig an LLM which outputs text), they are used as an intermediate step, which is then used to compute a simple
by danlenton 2y ago
They're not used as labels directly (we're not trainig an LLM which outputs text), they are used as an intermediate step, which is then used to compute a simple score which the neural score function is then trained on. The neural score function takes prompts and latent model representations in, and produces a score from 0-1
- iknownthing 2y agoTOS says you cannot "Use Output Data to develop models that compete with OpenAI." You're probably fine but it's close.
- danlenton 2y agoTo be honest we're thinking of moving away from this soon anyway. Open source models will soon make for perfectly good judges (or juries), with Llama3 etc.: https://arxiv.org/pdf/2404.18796 https://arxiv.org/pdf/2404.18796