For a while I was working on a browser for UI designers, which involved pretty heavy use of LLMs talking to CDP. In my experience, CDP is extremely token consumptive.
Haven't released it yet, but a little website for evaluating AI UI design. They're not actual evals, it's moreso just showing the results of different UI tasks across models, reasoning levels, skills, harnesses, etc.
Right. It feels like we're living in a parallel world or something. Can agents code? Sure. Can you let them code on their own for a serious production project? Not a chance.
Which serious production projects have AI agents coding on their own? And I’m assuming that means they are routinely taking tasks and deploying them to production autonomously
Right. The only way I'd feel comfortable doing that is if I spent an inordinate amount of time writing very detailed specs, so in terms of labor offset I'm not sure the juice would even be worth the squeeze there. In other words, a sufficiently detailed spec is nearly indistinguishable from code.
What people seem to be wanting is for an agent to infer vast complex data from terse simple data, which I think is probably impossible on a philosophical level. There's real information loss in language, and compute can only make guesses at the end of the day. I really don't see how we bridge that gap.
Astra does exactly the same sort of things that Sol or any of the previous agents do. They duplicate code, overengineer, miss the point, etc.
I was very optimistic about it when it was announced and saw all the demos, but a week later I find it only marginally better (and in some cases worse) than before.
Yeah, Astra, not like it matters. It will gleefully make db queries that perform full table scans instead of adding an index. I need to intuit that that's what's happening and then explicitly instruct it not to be negligently dumb.
Consider that the people who have never have never had to be on-call for a product that results in losses for the company in the event of an outage, and/or have never had to attend an RCA meeting have just as much of a voice in this discussion. Some segments of the software industry don't even have to worry about those sorts of things, and it's easy to get myopic after a while...I doubt that a long time dev at hobbyist-centric wearable manufacturers or audio plugin shops has the same sorts of concerns as someone who has specialized in ad delivery for social media sites.
Allowing only people who have served on call or made concrete contributions to multiple RCAs to have input on the software development process is a wonderful idea!
me too; and my coding agents are slowed down (from developing features) because I require them to refactor the code to be more readable; my code metrics tests force AIs to leverage functional programming and design patterns).
Yes agents can produce code that compiles and runs, but I had to add tools to keep them on track, document their work, follow a process, check their outputs. I also use other AIs to generate developer documentation and review code.
It is like managing a bunch of idiot savant eager-to-please interns, except unlike interns, coding agents do not (yet) learn and improve on their own.
I bought it on day one in 2016, and like everyone else, I was wowed by the planet hopping experience for about an hour or so. Then I put it down and didn't think about it again.
Then I picked it up a few years later, and discovered base building. Had more fun than the first playthrough, but it didn't click. Eventually I hear about something called the Galactic Hub, which is a star system marked by players as a central spot to build their bases. Kind of like a human home in the vast empty universe of NMS.
I get curious about it, and look up instructions on how to get there. I did some quick napkin math and realized it would take literally dozens of hours of grinding - nonstop travel - to get there. However, as it turns out, there's a loophole with black holes. When you enter a black hole, it spits you out into a totally different place in the galaxy. From the players perspective its random, but players got together and figured out that each unique black hole always sends you to the same spot in the galaxy. So they constructed a black hole "highway", along with an app that let you input your desired coordinates, and it would give you the list of black holes that would most optimally get you near your destination.
So I spend an evening playing the game, switching back and forth from the game to my laptop, inputting coordinates and slowly making my way closer and closer to the hub. Eventually I find it, and what was amazing was that it was the first time I'd seen any evidence of another player in the game.
That moment felt utterly magical, because while I was playing a game, I had an actual, real objective that I formed entirely on my own.
So yes, the game is an odd duck, but it can really open up with the right mindset.
That kind of play is pretty rare these days. The current way is to put in portal coordinates (which are pretty much mandatory in NMS social media) and just walk through to the destination. Or have someone take you there from the Anomaly. And while what you did is still possible and cool, would you have still done it if given the chance to take a portal? Would most players? (No.)
"We assume creative writing requires uniquely human qualities, such as emotional understanding and lived experience, which leads people to underestimate AI's capabilities."
It's not about the relative "creativity" of it. Art is fundamentally about communication. It's a social activity, whether you're reading a book or listening to music or looking at a painting. Even a mid-average author is going to make something more compelling, on a human level, than extraordinary prose generated by a machine.
AI-generated creative artifacts are as interesting as the fractals in a snowflake. Interesting, sometimes beautiful, but void of human meaning.
Is art what is generated, or is art what is in the eye of the beholder?
Do we find artistry in nature?
What does it say about us that the sum average of us disgusts us on a visceral artistic level, is that the machine's fault for reflecting us accurately?
Art is a social activity whereby a human uses their creativity to produce an artifact that expresses some aspect of their experience, using non-standard forms of communication. It's a way of signaling to others "do you see what I see? do you feel what I feel?" The beauty of art exists on two layers: the aesthetic quality of the work, and the sense of shared experience among those who consume the art.
We find aesthetic beauty in nature; we do not find artistry.
The most interesting creative decisions happen within a tight set of constraints.
Side note, I also had to expand my definition of "creativity" early in my career. I grew up and was educated as an artist. Even received a Bachelor of Fine Art Painting. Never took a single class on anything to do with computer science. As I was transitioning my career towards programming, I was worried that I would lose my creativity. It took me a while to realize that programming was intensely creative, just as much as being an artist. They simply have different aesthetics.
I've been on this kick for a long time. A few years ago I was exploring a UI design tool that was entirely keyboard-driven, using pneumonic keychords inspired by Vim: https://github.com/danielvaughn/stride
My experience is that while it's awesome to have really deep keyboard-driven experience, it can't _only_ be that. You need some graphical controls to help guide users.
To be fair, the article doesn't say that AC is banned - it mostly focuses on the HOA-style local ordinances and permits that discourage people from using AC.
I know someone who works at a data center for the IRS, and a lot of their records are all stored in PDFs. He thought it was disgusting when he first learned about it, but eventually realized that it's essentially resistant to a large scale data leak. Retrieval just isn't fast enough.
Presumably the pdfs are so large that moving any significant chunk of them over available bandwidth takes so much time, and requires so much storage, that doing so just isn't useful for the attacker.
While I don't know any of the relevant numbers involved, imagine a scenario where exfiltrating even 10% of the total data would take years. What would be the point?
A malicious actor can target individual records, but leaking a lot of them becomes impractical.
reply