I’ll leave it to other comments to discuss the societal and moral implications of being able to do this (which, I agree, ick…). On a practical level:
We train an encoding model, a “digital twin”, that predicts how each visual region responds to any video. Now we can ask: which video would make a chosen region light up the most? NEvo searches for that video automatically, using the twin’s prediction as its reward.
I only scanned the paper, so maybe I missed it, but is there any confirmation that this ‘digital twin’ works? Like, do the generated videos actually cause the same patterns as in the ‘digital twin’ brain model in real humans in an MRI machine? My instinct is to be skeptical that it’s possible to reliably create a video -> brain activation prediction model.
It kinda worries me that one of the big points here is that you can give non-technically-inclined users power by allowing them to build Home Assistant interfaces using Claude.
In my experience (and one look at the Home Assistant forums shows I’m not alone), allowing AI to write Home Assistant configs is a questionable activity.
There’s so much old, wrong information out there that it regularly uses (or tries to use) deprecated or removed features, and often even hallucinates config fields if you ask it to do something that is unsupported (and, of course, the user doesn’t know beforehand what’s supported). And you better have backups, because once it’s inserted unparsable nonsense or messed up the white space in the YAML config files, Home Assistant just won’t work.
Admittedly, my experience here is six months old, and I was talking to Claude not using an agent - but still…
I actually find it hard to read. Something about it is hard to follow, and I have to think very hard to keep the ‘story’ straight in my head. I keep having to re-read sections, which it not normal for me.
This doesn’t make much intuitive sense to me because shouldn’t output based on statistics have a very strong through line?
This is where I really first got lost:
“You can’t run a greenhouse without gas unless you can see the moments it slips — a fault, a silent fallback — and close them one at a time. The monitoring doesn’t undermine the gas-free promise. It’s what makes the promise keepable.”
I love that the top two (currently) replies to this claim that the Economist is ‘centric’ are one that’s arguing that it’s obviously too left wing for that label, and another arguing that it’s obviously too right-wing.
I’ll admit to being ‘out of gaming’ for that era (I’m a pandemic gaming-returner), but I wasn’t aware of that. Was his reach bigger than his official title?
Wikipedia doesn’t show him in an ‘overall leader’ role until 2014, a year into the relatively disastrous Xbox One era.
The (an?) incoherency with that is that this was happening at the same time as the ‘everything’s an Xbox’ strategy that saw them produce games for other platforms too.
I never understood the esteem Phil Spencer was held in, seemingly both by fans and industry insiders. I never understood Phil’s strategy.
Buy a plethora of studios. Pay an order of magnitude over the odds for the big ones - just to be sure you get them! ‘Rescue’ smaller ones of questionable financial value - as part of Xbox they’d somehow be successful enough to justify the price paid. Heavily manage the studio heads - but, uh, also give them total creative freedom - and allow them to make niche games. Sell hardware at a loss - but also make the games available on all platforms. Don’t allow any software that takes advantage of your most powerful hardware, because it also has to run on the other, less powerful console you are also selling. Also, the future is streaming! But, uh, maybe not!
Not just the strategy, but almost every aspect of the ‘strategy’, was incoherent - as current management is very close to outright saying.
Games are much cheaper in real terms (even in terms of percentage of wages/income, if you don’t believe that’s kept up with inflation generally) than they’ve ever been.
I think this is a big element. I’ve been feeling this a bit too. I’ve been a physical-game buyer in the past - usually saving money by buying used games.
But I’m not going to keep a collection of old consoles in my living room, so at this point it feels more likely that I will be able to actually play games further into the future if I buy digitally.
This is the opposite of what proponents of physical media heavily argue - indeed, many here are making it their only argument - but I suspect it’s how lots of ‘regular people’ that aren’t saturated in digital rights politics feel.
“The Once and Future King” is an odd book. I read it recently - as an adult - and I’m not sure I ‘enjoyed’ reading it. It has a lot of ‘childishness’, especially in the first 2/3, of a kind I’ve never really liked in it. Perhaps the kind _adults_ think kids enjoy. But it’s also full of stuff like this at unexpected moments. Wonder at the world; consideration of others; the burden of leadership.
I ended up thinking of it extremely fondly - way more fondly than I would’ve expected when halfway through. It’s one of my favorite books in spite of itself. I’d recommend it.
So, I fully recognize this is dumb, but humans (like me!) are dumb and lazy.
The ability to just get in your car and have the phone-powered interface right there without having to take your phone out of your pocket and mount it in the holder is something I didn’t ‘niceness’ of before experiencing it.
We bought a Honda Prologue last year and didn’t even test drive the built-on-the-same-platform GM cars (e.g. the almost-identical Chevy Blazer EV) because they didn’t have CarPlay.
People like us have got to be affecting GM’s sales by some measurable amount, right?
This makes it sound like Sherlock was named in response to Watson. It was the other way around.
Earlier versions of Mac OS had an app called ‘Sherlock’[^1] that could search local files and the web in a fairly rigid manner.
‘Watson’[^2] was a third party shareware app very much inspired by Sherlock (and obviously, given the name, not trying to hide that!) that was much more flexible, more ‘OS X-like’, arguably much more user friendly, and was open to plugins (like, there was a movie time search plugin, an eBay plugin, an Amazon plugin etc).
Sherlock 3[^3], in MacOS 10.2, was redesigned with a UI very like that of Watson, and also allowed similar plugins, making Watson obsolete.
In the Apple developer world, “being Sherlocked” came to mean “your app being made obsolete by Apple including identical functionality with the OS”.
We train an encoding model, a “digital twin”, that predicts how each visual region responds to any video. Now we can ask: which video would make a chosen region light up the most? NEvo searches for that video automatically, using the twin’s prediction as its reward.
I only scanned the paper, so maybe I missed it, but is there any confirmation that this ‘digital twin’ works? Like, do the generated videos actually cause the same patterns as in the ‘digital twin’ brain model in real humans in an MRI machine? My instinct is to be skeptical that it’s possible to reliably create a video -> brain activation prediction model.