Ken Forbus

Wangcheng Xu shows how a hybrid system incorporating CogSketch, a symbolic model of high-level human vision, with off-the-shelf deep learning vision components can be used for diagram understanding. Competitive with VLMs while providing inspectable representations.
www.cogsys.org/proceedings/...

I would never ship something with error rates like these. Have the vendors no pride?

Ditto. But at 70 you will feel better for having done it

Because, given their track record, very likely the only violence was in ICE fantasies. Trib just finally posted the real story about an unprovoked attack where a woman was shot multiple times. Video exonerated her.

Public pressure. Lawsuits. Government shutdown. Not much, but better than nothing.

What is happening with this? No IGs left to police the government?

What can be done?

50,001 :-)

I'm told by Star Wars fans that if the executives had watched one of their own shows, Andor, they might have shown some spine.

Because Disney+ does not respect the First Amendment, i.e. taking Jimmy Kimmel off the air on ABC when all he did was tell the (inconvenient) truth, I've cancelled it and Hulu, which they also own.

Sad indeed

Unfortunately apt

X aside, the Findings hack sounds like a good one

Wish you’d posted your note here I hate to use X even as a visitor…

Nice. One tactic I use is to look puzzled and quietly ask “you can see me?”. Plenty of time to move on.

Hot off the presses: A journal article on our Hybrid Primal Sketch research has just appeared in Cognitive Systems Research.

You can download it for free until 10/28/25 at
authors.elsevier.com/a/1lh9w4xrDw...
Enjoy!

It’s quite good.

Coming to an anthology soon, for non-subscribers?

This is so counterproductive. More people, fewer hours, more creativity

No chance to train on the test set, eh?

Great essay. We’re far from the point where AI systems have sufficient agency that these become practical matters — our work on formalizing social and moral norms could be ultimately abused into something like a governor module — but I think it’s important to do better.

The lead author, Constantine Nakos, should be finishing his thesis this summer and will be job-hunting. A great opportunity for any research organization interested in novel cognitive architectures.

The journal version of our paper, “Towards knowledge autonomy in the Companion cognitive architecture”, is now available. It describes how we handle provenance and incremental learning of knowledge.
Journal version: doi.org/10.1016/j.co...
ACS conference version: arxiv.org/pdf/2407.06401

It's been a great ride, hope there are subsequent seasons!

One of my recently graduated students, Taylor Olson, has a nice paper on using moral norms to protect against adversarial training

Alligator Auschwitz has also been suggested

Not a good sign

Promising: analogical learning over qualitative representations. Will Hancock’s PhD thesis includes a Webots tabletop domain, most of the earlier papers use Freeciv. Thesis should be out soon, but earlier papers at QRG.northwestern.edu

Tariff Analogy

The number of fingers generative AI draws on a person