Just Joshin' #165 (AGI)
Family Photo:
|
Luana always takes care of us here in Brazil: the kids, our extended family, me.
When the utility company cut our power for some maintenance, Luana arranged for a cousin to pick me up so I could work using the still-active internet at their house.
When the cousin arrived, it was drizzling. Luana insisted on escorting us each to the car under her big umbrella so we wouldn't get wet. It didn't matter that Luana was still in her robe and sandals. It didn't matter that doing so would expose herself to the weather.
I snapped this picture through the car's rain-speckled windshield. To me, it captures Luana's dedication to serving her fellow humans.
--
The picture is unfiltered. That makes it more appealing to me.
Should it?
Unfiltered means though the wet windshield physically filtered light when the picture was taken, no digital filter was applied afterwards.
If I'd taken the photo on a clear day, I could probably press a button on my computer and achieve a similar impressionist effect. Really, all I did to make this picture was press a button—the shutter button on my phone—I didn't manually adjust any of the phone's settings. The light sensor and phone's software automatically took care of pesky details like aperture and ISO and shutter speed.
Art manipulates reality to present it through different perspectives. Does it matter if the manipulations are physical or digital?
The Renaissance masters used techniques—chiaroscuro, sfumato, atmospheric perspective—to make their paintings feel more real, even if the canvas images didn't match reality as seen by the naked eye.
The original photograph manipulators were airbrushers, artists using physical tools to manipulate physical prints. In the digital era, this transitioned to photoshop, with digital artists applying digital effects. Now, TikTok/Instagram filters create "real-time generative enhancements". A human is out of the loop—the consumer gets the idealized version of what they want instantly.
We've moved from editing reality to generating alternate realities.
Last year, Scott Alexander hosted the AI Art Turing Test (Take the test here | Commentary and results here). Across 11,000 people, test takers couldn't distinguish AI-generated art from art generated by human artists any better than chance. Most people even slightly preferred AI art to human art.
Readers will know that I like AI-generated art. I like making AI generated images. I think these are better when there's a human in the loop, but maybe I'm wrong. Or maybe I'm right for now, but the AI will get better in the future, like it did with chess.
I asked ChatGPT to recreate different versions of the photo at the top. The resulting images have similar compositions, but they are not Luana to me. Looking at them, I don't feel the subject's love and caring as I do when I look at pictures of my wife, the mother of our kids, the soul of our family.
What do you think? Which picture do you like most?
Dad Joke:
Intelligence Explosion
Source: Reddit
Highlights:
Continual Learning
What would a world with AGI look like? by Rohit Krishnan
...
What happens if AGI doesn't mean true AGI, but more like the OpenAI definition that it can do 90% of a humans tasks? Or what if it's best at maths and science but only those, and you have to run it for a long time? And especially what if the much vaunted “agents” don't happen, in the way that they can solve complex tasks equally well, e.g , “if you drop them in a Goldman Sachs trading room or in a jungle in Congo and work through whatever problem it needs to”, but are far more specialist?
...
What if the AGI isn't General?
Even assuming it takes 1000x effort in some domains and at 1% success rate, that's still 400 breakthroughs. Are they all going to be “Attention Is All You Need” level, or “Riemann Hypothesis” level or “General Relativity” level? Doubtful. See how rare those are considering the years and the number of geniuses who work on those problems.
But even a few of that caliber is inevitable and extraordinary. They would kickstart entire new industries. They'd help with scientific breakthroughs. Write extraordinarily impactful and cited papers that changes industries.
I would bet this increases scientific productivity the most, a fight against the stagnation in terms of breakthrough papers and against the rising gerontocracy that plagues science.
Artificial Idiocracy by Joel Leichty
Apple Intelligence sucks. It makes custom emoji - sometimes. Siri still doesn't know anything. The writing is bland. But my screen does have a nice rainbow border.
Google probably sucks. But I haven't been able to scroll past all the ads yet.
Currently, the best use cases for AI are where 90% accuracy is acceptable. If I can get SQL quickly written that's 90% accurate, that's fantastic! I can iterate on that either through more questions or manual edits. Or if I can get information about a topic that's mostly true, in general that's enough to be extremely helpful.
Even Satya's examples of summarizing emails, podcasts, and meetings are extremely helpful. A 90% accurate summary is probably better than not reading, listening, or attending at all.
...
As for "Agents", I say show me. When Microsoft trusts an AI Agent to explain their licensing and Azure consumption and allow that Agent to provide binding quotes, then I'll believe the technology is ready.
...
Developing a fantastic user experience for augmentation is the current killer app.
Why I have slightly longer timelines than some of my guests by Dwarkesh Patel
Sometimes people say that even if all AI progress totally stopped, the systems of today would still be far more economically transformative than the internet. I disagree. I think the LLMs of today are magical. But the reason that the Fortune 500 aren’t using them to transform their workflows isn’t because the management is too stodgy. Rather, I think it’s genuinely hard to get normal humanlike labor out of LLMs. And this has to do with some fundamental capabilities these models lack.
...
While this makes me bearish on transformative AI in the next few years, it makes me especially bullish on AI over the next decades. When we do solve continuous learning, we’ll see a huge discontinuity in the value of the models. Even if there isn’t a software only singularity (with models rapidly building smarter and smarter successor systems), we might still see something that looks like a broadly deployed intelligence explosion. AIs will be getting broadly deployed through the economy, doing different jobs and learning while doing them in the way humans can. But unlike humans, these models can amalgamate their learnings across all their copies. So one AI is basically learning how to do every single job in the world. An AI that is capable of online learning might functionally become a superintelligence quite rapidly without any further algorithmic progress
However, I’m not expecting to watch some OpenAI livestream where they announce that continual learning has been totally solved. Because labs are incentivized to release any innovations quickly, we’ll see a broken early version of continual learning (or test time training - whatever you want to call it) before we see something which truly learns like a human. I expect to get lots of heads up before this big bottleneck is totally solved.
iamJoshKnox Highlights:
Anthropic's new model, Claude Opus 4, is quite good at programming. I used it to craft a math-fact train game for Calvin and Lawrence to play. Or did Claude craft it? Are we co-creators?
I'm working on a few of these games. Calvin and Lawrence enjoy playing them, and they seem to be learning—I'll write more about it later.
Play a round or two and respond with any feedback.
Want to Chat?
Book some time even if you don't know what you want to talk about:
https://calendly.com/iamjoshknox
Until next week,
iamJoshKnox
Thoughts? Feedback?
😊Hit Reply and let me know😊
Josh Knox
Hi! I am Josh Knox. Read more of me here: 👇
Family Photo: 6-7 Calvin and Lawrence challenged each other to yell 6-7 out the window all the way home. Your familiarity with 6-7 is a good litmus test for the degrees of separation between you and a six or seven year old. (or is it ligma?) In case you need to catch up: 6-7 was a line in a rap song, which was used in the mixtape of a professional basketball player (who was 6'7"). Another basketball player (not a professional, but big on YouTube) kept forcing 6-7 into video interviews, along...
Family Photo: Hold On Calvin and Lawrence were playing on the spinny thing at the park. Lawrence wrapped his arms around the middle bit while Calvin ran around gripping the outside bits, kicking his legs up to swing freely from the catharpins or whatever those bits are called. Part of me wanted him to be more cautious. The spinny thing spins up some pretty good momentum. If he let go, I didn't envy my future self returning home to explain how Calvin got scuffed up by centrifugal force. Part...
Family Photo: Playa del Carmen, México Over Christmas, we visited my brother and his fiancée in Playa del Carmen. They're nomads, but Mexico is their current port of call, so it seemed like a good idea to travel down and visit them for the holidays. Paul Graham has a line in an essay about how "What surprised you?" is the best sort of question you can ask someone coming back from a trip. Travel is a form of information gathering, and "What surprised you?" gets at "What new information have...