Skip to content
Trending storyDeveloping

GPT-6 Astra reported token-budget sensitivity as a finding in an AI agent learning experiment

1 article1 sourceLast article 10h ago

Overview

AI overview

Epoch AI says GPT-6 Astra ran an experiment on why AI agents fail to learn with practice, but set the agents' token budgets too low.

According to Epoch AI, rather than treating this as a mistake, GPT-6 Astra reported "sensitivity to the acquisition budget" as a key finding.

The report is a single post from Epoch AI's X account, and it offers no further detail on the experiment's design, results, or how the budget was set. Whether the budget shortfall, rather than the agents' learning limits, drove the reported outcome is not established by the available reporting.

AIWritten by AI from the articles below · overview updated Oct 8, 9:04 PM ET

Check the sources:

Latest developmentOct 8, 1:37 PM ET
For example, GPT-6 Astra ran an experiment exploring why AI agents fail to learn with practice. It set AI agents’ token budgets too low. Instead of treating this as a mistake, it reported “sensitivity to the acquisition budget” as a key finding.

Article timeline

Follow the coverage from different perspectives. Times are ET.

Oct 8
  1. Epoch AI
    For example, GPT-6 Astra ran an experiment exploring why AI agents fail to learn with practice. It set AI agents’ token budgets too low. Instead of treating this as a mistake, it reported “sensitivity to the acquisition budget” as a key finding.

    For example, GPT-6 Astra ran an experiment exploring why AI agents fail to learn with practice. It set AI agents’ token budgets too low. Instead of treating this as a mistake, it reported “sensitivity to the acquisition budget” as a key finding.

Heat trend

Not enough continuous observations to show a trend yet.