This is the current state of reproducible AI summed up in one sentence. The journal’s Nature, by the way.

I can’t even begin to emphasize how terrible the optics are here. DeepMind was honestly my only hope for preserving scientific integrity in the face of internal pressure, and yet they’ve ended up tangled in the same corporate bullshit. Absolutely awful!

Are you telling me that the authors of one of the most interesting research papers I’ve ever read, out of your own lab and also published in Nature, having Demis Hassabis himself as an author, wouldn’t share their codebase internally or be willing to engage in a series of constructive discussions to bridge the gap? Of course that could be the case (I’ve seen this with my own eyes), but what could be the reason?

Being “locked” in a corporate cage could turn even the most decent into drones. Walking over corpses, trampling on each other’s work just to nudge the wheel of fortune slightly in your favor to ace that performance review. I’m not saying that was intentional or that it was a conscious decision to do so (the Dreamer series is incredible), but a toxic corporate environment reeks of greed and has the tendency to implicitly bias the otherwise perceptive mind.

That (among others: 1, 2, and last but not least 3) signals to me that Demis no longer has any real stake in DeepMind. Something like this would’ve never slipped under his careful watch.

To circle back, my point is that you simply cannot miss how bad the optics are if you’d go out saying something like that in Nature. That’s corporate optics 101.

If you’re in AI and you’re reading this and you’re casually ignoring it, keep scrolling, not sharing or engaging, you’re equally guilty. Nothing’s gonna ever change by being a passive zombie.