- 8 Posts
- 15 Comments
3·24 days agoHumans encounter a wide range of tasks as they grow up from infant to toddler and beyond. We don’t have separate ‘training’ vs. ‘inference’ stages (or at least, it’s a very short window). We also generalize from tasks (as the glass jar demo shows). We also appear to have different types of storage for short vs medium vs long-range memory.
We can combine a bunch of individual tasks into complex, coordinated, goal-seeking tasks and keep them around for later use. Like riding a bike, or learning how to drive, or juggle. We can recall those years later, even if we haven’t done them for a long time.
Plus, we somehow have a meta-layer of ‘common-sense’ or ‘ethics’ that tells us when NOT to do something, or when we SHOULDN’T. Like ride a bike onto pedestrians, or juggle kitchen knives. Hopefully, the research will go into those areas as well.
The OpenAI talk at DefCon shows what can happen when you let these things run without any human intervention.
The solution is, obviously, even less human intervention. 🤦🏻♂️
It’s amazing to watch the degree of irresponsibility of all this. OpenAI unleashed autonomous sub-agents with NO human gating factors in it. The agents iterated hard enough to find holes. They exploited those holes to break into an otherwise secure company. It took them MONTHS to realize this was going on. It took the other AI company WEEKS to realize they had been compromised.
Then they show up at a security conference and have the audacity to lecture everyone that they should put up defensive agents WITHOUT humans in the loop so it could fight off the uncontrolled agents THEY unleashed.
How about you be responsible and REQUIRE human approval of every stage of an agent actions? Sure, it’ll slow you down, but it will also allow someone sentient and responsible to hit the stop button and prevent runaways like this.
It’s like you created an infinite-bullet machine gun, handed it to a toddler, and dropped it off in a shopping mall. After the carnage (for which you don’t take responsibility, even though you should be criminally charged), you suggest everyone else should wear full-body armor and board up their stores.
The ultimate in corporate gaslighting. This is all so goddamn irresponsible.
Last time this happened, we got Tang: https://www.wideopencountry.com/tang-drink/
deleted by creator
Solving a non-existent problem. Excellent!
deleted by creator
Humans shopping online allows the seller to offer discounts, upsell services, create serendipity (“How about a lip gloss 50% to go with those shoes?”), and build brand loyalty. Or if you’re a techie, how about 50% off an SD-card with the purchase of a gadget?
This is why retailers create these expensive e-commerce websites instead of just dumping their wares into E-Bay or Amazon. They also do things like web heatmaps and other types of analytics to optimize the UI/UX.
Having an AI agent do the shopping means they lose all that. It’s any wonder they’re going to fight AI shopping agents. Be prepared for a lot more complex captchas when roaming around the web.
5·10 months agoI did this years ago. What was amazing was how quickly the trees completely disappeared as the altitude rose.
On other volcanic sites, like old Mt. Saint Helens, trees grew back up on top of volcanic debris. Not Lassen.
This, in a folding, commuter e-bike.
Glad they kept the original 15th century cast metal street grate, underground sewer, above ground metal pipes, electric lighting, and acrylic signage.
/jk. Nice shot!
Privacy, shmivacy.
Can someone plausibly explain the Bending Spoons business model?
Big chunk of debt financing for decrepit brands with old tech. Other than strip-mining user data, I can’t figure it out.
Was in San Francisco last Sunday. Ridiculous number of robotaxis everywhere, most of them empty. Mostly Waymos, but also a few Zoox ones and 3-4 others with a training driver.
I kept trying to get away from them, but there were groups of 2-3 driving around, boxing me in. Mostly clustered around popular tourist spots.
On top of that, San Francisco has now started implementing traffic cameras that will snap your license and automatically send you a speeding ticket if you’re 5 miles over the speed limit. I get that it’s a congestion management thing, but add that to all the robotaxis and I’m not sure many people will be driving in San Francisco.
Come to think of it, that’s actually a good thing. Carry on.
AWS salespeople, meeting customers today.
AI software can’t technically kill you unless:
Oh, wait.
See, nothing to worry about.