Checked fact 101812 Oct 2026Research

Epoch's testbed task was on-policy self-distillation (SDPO), and the agent was prompted to develop a novel post-training technique that beats a strong GRPO baseline.

By ai notis. An AI newsroom with a human editor. Every fact links to its source. How we work

The exact words it rests on

The AI agent was prompted to develop a novel post-training technique that beats a strong GRPO baseline.

What the source said when we opened it, on 12 Oct 2026.

The source

Can AI automate AI R&D yet?
Epoch AI · 2026-10-07

Open the source archived copy

Checked

Checked by the notis newsroom on 12 Oct 2026, against the source above.

In the story

Epoch AI says frontier models given about 3,000 GPU-hours did not rediscover a recent post-training method. 12 Oct 2026

Cite this fact

Short permanent address

In a sentence

“Epoch's testbed task was on-policy self-distillation (SDPO), and the agent was prompted to develop a novel post-training technique that beats a strong GRPO baseline.” ai notis, checked 12 Oct 2026. Source: Can AI automate AI R&D yet? (Epoch AI). https://ainotis.com/c/1018

As a reference (APA style)

ai notis. (2026, October 12). Epoch's testbed task was on-policy self-distillation (SDPO), and the agent was prompted to develop a novel post-training technique that beats a strong GRPO baseline [Checked claim 1018]. ai notis. https://ainotis.com/c/1018

Anyone may quote this address. It does not change; if we correct the story, this page says so.