Inside How Anthropic Is Building the Next Claude | Alex Albert

Channel: Peter Yang

Uploaded:

Duration: PT2103S

Read a transcript preview of “Inside How Anthropic Is Building the Next Claude | Alex Albert” from Peter Yang, uploaded on 2026-05-17.

Transcript preview

Language: English (original, YouTube automatic captions)

[0:00] I was definitely the first prompt

[0:01] engineer at Anthropic. I might have been

[0:02] the first in the world. We treat the

[0:04] model as if it's a product to some

[0:05] degree. With every new model, we are

[0:07] specking out exactly what do we want

[0:09] this model to be good at. When the agent

[0:10] isn't running a task for you, or maybe

[0:12] it's in the background, it's actually

[0:13] going through its memories, finding

[0:15] things that [music] might contradict,

[0:16] pruning them, cleaning them up. This

[0:18] concept of dreaming. Engine time isn't

[0:21] as much of a one-way door these days. If

[0:22] it's something that's not a one-way

[0:24] door, that's effectively free at this

[0:26] point. Agents that are running on task

[0:28] for a long amount of time and they're

[0:29] having to make a lot [music] of judgment

[0:30] decisions. The questions of what its

[0:32] character is and what it cares about are

[0:34] very important. Did you have to Did you

[0:35] have to try to avoid consciousness when

[0:36] you were training it and stuff? This is

[0:39] a big question.

[0:43] Alex, it's great to have you here today

[0:45] with Clock Conference. And you used to

[0:48] be head of DevRel at Anthropic, right?

[0:49] And you recently became product manager

[0:51] for the research team, right? Is that

[0:53] it? So,

[0:54] I've been a PM for like over a decade

[0:55] now, too. As a PM, you kind of try to

[0:57] understand the user problem, you try to

[0:59] identify solution, you try to build

[1:00] stuff. But, you know, I have no clue

[1:02] about how a PM works on a research team.

[1:03] Maybe we can talk a little bit about

[1:05] that.

[1:05] >> It's very similar in that sense.

[1:07] Um

[1:08] I'm always wanting to talk to customers,

[1:09] get as

Sign up free to read and search the full transcript