3 posts tagged #Frontier research.
A new preprint finds that AI agents with memory sometimes fail tasks they would solve without it, by following a wrong lesson. Its fix, WikiTrace, links every lesson back to the run it came from so the agent can check it.
A new study says yes: poisoning 0.1% of a model's pretraining data planted a backdoor that made an AI agent run an attacker's command, and it survived clean training afterward. Here is what the researchers found.
On November 18, 2026, NASA's Voyager 1 becomes the first human-made object to reach one light-day from Earth. Here is what that distance means and how the 49-year-old probe is holding up.
