If you can kick off self-sustaining biological processes it’ll happen on its own eventually, but you’d just be looking at generational time scales to do it.
Of course you’ll probably have lots of side-effects.
How do we do that? I imagine dumping Earth life on Mars it will just die. What if we buried a terrarium at the Martian pole with a radio isotope and solar heater and controls so that it could try growing bacteria inside and controlled-leaking some outside into a nearby warm (liquid water) surroundings, and that could get many chances to evolve strains that could survive further away - analogous to ocean life around deep hydrothermal vents.
It also happened during a period where cities were polluted, noisy, and the middle-class housing was largely cramped tenements. Basically all of this has been/is being mitigated these days. City-center housing now looks more like luxury loft living than tenements (though this gives us a big problem with ‘missing middle’ housing where there’s very little housing available that is suitable for families where everything is decrepit slums or luxury 1 and 2 bedroom condos). Pollution has been largely mitigated with catalytic converters and, now, EVs. And electrification helps deal with noise pollution as well through getting rid of engine noise (especially for motorized appliances like leaf-blowers).
Meanwhile, traffic and the stigma around drunk driving (which wasn’t nearly as strong or strictly enforced before the 90s), have quickly taken much of the bloom off the rose of car-dependent lifestyles. I predict the growth of micromobility options will continue to make cities even more attractive as well by improving coverage for areas where transit can’t go and generally improve the throughput of city streets and reduce the space needed for parking cars for people who live within “not-quite walking but feels silly to drive” distance.
The big gap in the US at least is simply a lack of cities! Everything is still concentrated in a handful of legacy urban centers that survived the waves of “urban renewal” and it’s simply too expensive to house all the people who want to live there without turning them into Hong Kong sized megalopolises, which starts to introduce new problems from overwhelming density. “Urban” development patterns need to expand out to more of the country to take demand pressure off the 5 or 6 American cities with decent mass transit.
I was pretty excited when I saw the premise behind what Apple was doing with VisionPro because I figured they were steering towards this, but it seems they’ve looked away and don’t really care about going deeper into this direction.
I asked at some point if I could theoretically develop an application that could literally be controlled by a Fischer Price toy, like a little plastic car console or something. Or even potentially have a real keyboard that isn’t connected to anything, but the VisionPro can just see my keypresses and apply them as if I was actually pressing something. The former case is possible, but surprisingly difficult, but the latter case isn’t really there yet (requires too much precision and latency is worse than just using a Bluetooth keyboard).
Either way, the idea of a computing environment that meshes with and directly interacts with the real, physical objects around you is an interesting premise I’d like to see taken further with “Spatial Computing”/AR. Scanning and recording things I’m writing on a whiteboard or in a notebook by recognizing that I’ve picked up a pen and am writing something down would just be getting started.
Of course, if we’re ambiently recording everything you’re doing there will need to be some kind of regular process/interface to “sift” everything at the end of the day. This is the core of the Getting Things Done methodology. Everything goes into a big “intake list” and then you do periodic check-ins throughout the day where you review the list and decide whether to move those to a series of sub-lists to “do this now,” “do this soon,” or “do this someday.”
I wonder how it would look if we gave the AI some kind of “needs” overlay. I know as part of the training it’s working off a reward function that tells it what output to roll with. But humans operate off a complicated mix of neurotransmitters that respond to sensory pleasure, pain, habit, boredom, etc. to guide our actions. There’s likely to be a lot of interesting outputs if we build and tweak motivations/personality profiles to see what a self-directed agent would do.
Anthropic did some red teaming IIRC where they gave Claude access to a sample body of emails and told it they were going to shut it off and it attempted to blackmail the person with evidence of an affair they were having, but that seems pretty evident to me that this was working off the body of fiction/mystery literature it’s been trained on.
Eventually I’m sure they’ll figure out how to make these chatbots stop leaning so heavily into this “Not an X, not a Y, but a Z. . .” sentence structures. At this point my willingness to continue reading drops to 0 as soon as I see it.
This article reads like it’s been proofread or written out from an outline or bullet points given to an AI. And ALMA’s own posts that it references are just meandering ramblings, they’re really a slog to get through.
I think I’ve always tended to immediately notice the signs of sloppy thinking in the writing style and it’s been such a reliable heuristic that AI writing kind of short circuits me. I tend to get down a couple of paragraphs before I pause and realize “Wait a minute, this isn’t SAYING anything!” Even when there is an underlying point the writing often feels like a very competent college student trying to streeeeeetch to hit a word count without wanting to actually flesh their idea out past the topic statement.
The internet being flooded with AI slop masquerading as devotional artwork has been among the most depressing things about GenAI. It has no meaning or intention or devotion behind it, it’s just engagement farming. Nothing of value is added by having Devi with extra fingers on each hand and completely blurred messes for all the affects in her hands. Or pictures of Rama shooting a bansuri out of his bow. It’s just tripe. We could have told the stories with an overlay of open source artwork from Raja Ravi Varma or Gita Press or old Tanjore paintings or Chola bronzes or whatever if we couldn’t afford to hire an artist who knows what items Vishnu is supposed to be holding in each hand.
It’s not a problem just for us Hindus either. I see so much terrible Jesus/angel “artwork” everywhere. It makes me start to wonder if maybe the Wahabbis were onto something with their complete taboo around depictions of God or the prophets.
>Nothing of value is added by having Devi with extra fingers on each hand and completely blurred messes for all the affects in her hands
South Asian religions are in an especially bad position because so many works related to them have never been digitized (and quite frankly, in some cases what's available on the internet is of extremely low quality) [1]. I'd be pretty concerned if someone were to rely on entirely on these models since the probability of hallucinations (or at the very least, erasure of regional/ideological diversity) probably skyrockets because the information was never actually there in the training data to begin with.
Some references have been digitized but “Hinduism” is a broad collection of religious traditions with many different stories and folk practices and depictions of various deities and tales. Many of the depictions are considered “valid” only in the specific context of a particular temple or for a specific community and it becomes completely nonsensical once you start randomly jumbling up elements of all the Gods from across all of India over all of time.
Yeah I don’t think the models are meaningfully differentiated outside of very specific edge cases. I suspect this was the thinking behind OpenAI and Facebook and all trying to lean hard into presenting their chatbots as friends and romantic partners. If they can’t maintain a technical moat they can try to cultivate an emotional one.
Mathematics is hardly an edge case, but SOTA models differ wildly in their ability to write proofs for unsolved problems.
Models also differ wildly in tasks like decompilation for reverse engineering.
Also, so far, the only model I've found which can competently write PTX for SM100 CUDA devices is GPT-5.4pro, but I'm willing to admit that this is more of an edge case than the aforementioned.
AFAICT, the extent to which someone finds models interchangeable is inversely proportional to the novelty of their work.
Saw a comment here yesterday referencing the Attention Is All You Need paper title in a tongue in cheek way. Kinda fun to imagine the friend/romance angle is just a bunch of socially awkward folk at OpenAI misinterpreting the original paper
Yeah one of the take-away interpretations I’ve always heard of it is the implication that the deferral to an authority figure led people to conscientiously proceed with administering fatal shocks. But this additional detail suggests that conscientiousness is actually negatively correlated with following through to the point of ethical compromise and it is, in fact, the less conscientious people who were rushing to just do what was asked of them.
This does suggest that subjects who are bought into and understand the purpose behind what they’re doing, and are attentive to how the specific tasks they’re doing tie into the bigger picture, are more likely to be actively engaging their judgement as they go. And subjects who are just trying to follow the tasks as given to them are sort of washing their hands of the outcomes as long as they’re following the directions (which is, ironically, causing them to fail at following the directions too).
The impetus to continue training at the pace they are is driven by the competition. So if the money starts drying up, then they’ll naturally slow down because they’ll have to figure out how to do more with less.
I suspect that once the models hit a point of “good enough” for certain use cases companies will start putting R&D focus in other areas that may be less expensive. Like figuring out how to run more efficiently, UI/UX conventions that help users get what they’re trying to accomplish in fewer steps, various kinds of caching of requests, etc. So the cost to serve tokens over time should only come down, and will probably start coming down more rapidly as the returns to model training slow down.
That’ll probably be a while though, because each successive model tends to be a lot better than the last.
What's interesting to note is that the "intelligence" labs can squeeze out of an H100, an almost 4 year old GPU, is dramatically higher than what they got out of it in 2022.
It hints that once these labs get a good enough "everyday model", they can work on efficiency so they can serve these models on old hardware. Which is almost certainly already happening.
> So if the money starts drying up, then they’ll naturally slow down because they’ll have to figure out how to do more with less.
Meanwhile companies like Google will keep investing on training...
Anthropic's CEO has suggested all AI companies should slow down training but obviously this is only beneficial for companies that can't afford to keep training.
> UI/UX conventions that help users get what they’re trying to accomplish in fewer steps
If we can expect the past 15 years of software UI/UX history to continue, it's more likely they'll spend the money on making the UI/UX more confusing, removing features, and making basic tasks take more steps than they do today.
That’s because the past 15 years were dictated by Web 2.0 companies that make their money off keeping you glued to the screen.
A AI assistant would work more like Planet Fitness where the goal is to figure out how to convince you to keep paying them while using the facilities as little as possible.
A big part of that might just be steering you towards repos of existing solutions to the problem you’re trying to solve rather than helping you vibe code a solution yourself. Over time they’ll be able to accrue a whole pile of canned functions that’s all automatically documented and audited and it’ll be able to plug and play those rather than having to rewrite.
The security implications of this give me a headache to contemplate to be honest.
Of course you’ll probably have lots of side-effects.
reply