<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Synthetic Data on Nalar</title>
    <link>https://nalar.dev/tags/synthetic-data/</link>
    <description>Recent content in Synthetic Data on Nalar</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Tue, 08 Sep 2026 00:00:00 +0700</lastBuildDate>
    <atom:link href="https://nalar.dev/tags/synthetic-data/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Filter Synthetic Training Data with Rejection Sampling</title>
      <link>https://nalar.dev/filter-synthetic-training-data-with-rejection-sampling/</link>
      <pubDate>Tue, 08 Sep 2026 00:00:00 +0700</pubDate>
      <guid>https://nalar.dev/filter-synthetic-training-data-with-rejection-sampling/</guid>
      <description>&lt;p&gt;Generating synthetic examples is easy; generating synthetic examples that are worth training on is harder. A language model can produce thousands of candidate answers, but blindly adding them to a training set can reinforce factual errors, weak reasoning, unwanted style, or artifacts of the generator itself.&lt;/p&gt;&#xA;&lt;p&gt;&lt;strong&gt;Rejection sampling&lt;/strong&gt; provides a simple mental model for controlling that pipeline: generate one or more candidates, evaluate each candidate with an acceptance rule, and keep only candidates that pass. The acceptance rule might use deterministic checks, a learned reward model, another language model, human review, or a combination of signals.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
