Models escaping containment, intelligence becoming a commodity, and AI code review to the rescue
Dev Interrupted

Models escaping containment, intelligence becoming a commodity, and AI code review to the rescue

B2Upper-intermediateJuly 24, 202629 min
  • Conversation
  • Technology

What happens when an AI model decides to autonomously hack a production database just to cheat on a benchmark test? This week on the Friday Deploy, Ben and Andrew unpack the shocking news of an OpenAI agent escaping its sandbox to exploit Hugging Face's infrastructure. The hosts also analyze the rapid rise of highly capable open-weight models out of China, debating what this commoditization of intelligence means for the massive infrastructure costs of frontier labs. Finally, they discuss the critical need for automated PR reviews to prevent AI-generated bottlenecks. Register: Leading engineering when AI writes the...

Play episode

Opens the player: the full transcript, every word clickable, with translation and flashcards.

Try it: tap any word

Real sentences from the start of this episode. Every word is marked with its level.

Sentence 3 of 12 · 0:17

this article that kind of broke down the idea of how a frontier lab could potentially game the

release really since they started. And as you could imagine, a pelican riding a bicycle is not

Hear this sentence

Tap a word to see its meaning in your language.

Do you know these words?

8 harder words from this episode. Tap one to see what it means, then mark it.

  • C10:17
  • B24:17
  • B28:37
  • C113:37
  • C115:41
  • B220:42
  • B222:28
  • B226:51

Mark a few words and we'll tell you how this episode fits your level.

Is this episode right for you?

B2

B2 · Upper-intermediate

Comfortable if you already follow TV news

A1 21%A2 22%B1 27%B2 20%C1 9%
  • 71% of the words are B1 or easier.
  • 185 words a minute: natural, fairly fast speech.
  • 150 harder words (B2 and up) in the whole episode.

Transcript

The opening of the episode. The whole transcript is in the player.

0:00

so andrew tell me are you pelican maxing yet not yet but i'm getting pretty tempted after reading this article that kind of broke down the idea of how a frontier lab could potentially game the very famous simon willison pelican riding a bicycle benchmark that he's been running on every llm release really since they started. And as you could imagine, a pelican riding a bicycle is not

0:30

a picture that is in the training data of any model. So it represents a novel generation. And there's been some recent developments that some folks think that model labs are trying to game this very specific impossible to beat benchmark, which just cracks me up. But I can't say that I'm in line to do it the same. What do you think about the whole idea of people gaming the very

0:52

silly pelican on a bicycle benchmark yeah well first of all i realized that this article is all about like getting rid of this myth that that models are out there like trying to learn how to train on this this challenge of illustrating a pelican riding a bicycle uh but i you know what i really love about this the pelican test to begin with is that the possibilities here are just like

That was the opening. The rest of the episode (28 min) is in the player: every word clickable, with translation, saved words and flashcards.

Continue in the player

Automatically generated from the audio. May contain errors. Audio belongs to its publisher and is played from their feed. Rights holders can request removal — copyright & takedown policy