This is just a random thought, but have you tried doing an 'agentic' pelican?
As in have the model consider its generated SVG, and gradually refine it, using its knowledge of the relative positions and proportions of the shapes generated, and have it spin for a while, and hopefully the end result will be better than just oneshotting it.
Or maybe going even one step further - most modern models have tool use and image recognition capabilities - what if you have it generate an SVG (or parts/layers of it, as per the model's discretion) and feed it back to itself via image recognition, and then improve on the result.
I think it'd be interesting to see, as for a lot of models, their oneshot capability in coding is not necessarily corellated with their in-harness ability, the latter which really matters.
I tried that for the GPT-5 launch - a self-improving loop that renders the SVG, looks at it and tries again - and the results were surprisingly disappointing.
I should try it again with the more recent models.
From what I can tell, its more about cashflow - basically companies need to spend most of their revenue or be taxed on it - and you can buy only so many servers.
Now capital can flow towards AI - I'm sure the reason why engineers at Boeing or GM don't make the same money as software devs do is that their industries are otherwise capital intensive, among other things.
Why are people so focused on India when it comes to outsourcing?
US dev salaries are so much higher than the rest of the world that basically you could hire anywhere in Europe and still save most of the cost per person.
You could go to LATAM if you want the same timezone.
On the corollary, salaries of capable Indian developers have certainly caught up to most Western countries, so that you wont be saving much per person.
> Social media is an extremely competitive landscape, with competitors rising overnight.
This is not true at all. There are two players. FB/Instagram and TikTok. Using one does not preclude using the other. Other than tiktok, who was the last new player in social?
> Google, Apple and Microsoft dominate the world with their products and platforms. Facebook & WhatsApp certainly doesnt.
There are all kinds of social media, its segmented by userbase, and culture/geography.
Telegram has 1B users (which is surprising to me, I thought it was an ex-Soviet thing), and there are entire geographic strongholds, such as Russia and China.
Russia and China still use iPhones and Windows, but entirely skipped out on Facebook and Whatsapp.
And for all the scorn it gets on HN, Facebook still works for some of my use cases: high school friends, low-contact relatives, obscure geography groups, the Philippines.
> There are all kinds of social media, its segmented by userbase, and culture/geography.
Your own link has Meta with 3 of the top 4 platforms. Can you really see any of the competitors overtaking them in even the medium term?
> Russia and China still use iPhones and Windows, but entirely skipped out on Facebook and Whatsapp.
China doesn't use Google either, and while they might use Windows they're staying off Azure which is where Microsoft's main business is these days.
Yes there are countries which stay off Meta. But they are just as embedded in the workings of the world as any of the companies you mentioned, probably more so. Government decisions are made by people using a mix of Apple, Google and Microsoft hardware - but all of them are communicating over WhatsApp.
This is something I have observed too. I think one of the ways things have gotten worse is generally people are less social, we're interacting with each other via screens, not in person.
While one could meme about being introverts, I just feel like the writing in older media, movies and other records of the time makes me feel people back then were just more comfortable with each other, more practiced in natural social interaction, and this lack of understanding has not only made modern media less compelling, the fact that we don't understand in general what people are really like has been a detriment to the fabric of society.
It's weird how people pep talk the AI - if my Jira tickets looked like this, I would throw a fit.
I guess these people think they have special prompt engineering skills, and doing it like this is better than giving the AI a dry list of requirements (fwiw, they might be even right)
It’s not surprising to me that the same crowd that cheers for the demise of software engineering skills invented its own notion of AI prompting skills.
Too bad they can veer sharply into cringe territory pretty fast: “as an accomplished Senior Principal Engineer at a FAANG with 22 years of experience, create a todo list app.” It’s like interactive fanfiction.
This remind me of so called "optimization" hacks that people keep applying years after their languages get improved to make them unnecessary or even harmful.
Maybe at one point it helped to write prompts in this weird way, but with all the progress going on both in the models and the harness if it's not obsolete yet it will soon be. Just crufts that consumes tokens and fills the context window for nothing.
If that's true, I wonder if this is a deliberate decision by cloud providers to push users towards microservice architectures with proprietary cloud storage like S3, so you can't do on-machine dbs even for simple servers.
It's probably a combination of high density storage nodes getting I/O bound and SSDs having finite write endurance. Anything that improves the first problem costs them money to improve it and then makes the second problem worse, and the second one costs them money again, so why would they want to make the default something that costs then more twice if most people don't need it?
Instead they make the default "meager IOPS" and then charge more to the people who need more.
I'm not sure about this but I remember that a lot of servers at my old company stuck with hard disks as late as 2018 - exactly for the same reason - HDDs for all their faults dont have write endurance issues. This was quite surprising to me back then.
How often is the storage in cloud providers even local vs how often are laptops doing anything other than raw access to a single local disk with a basic FS?
I remember my worked laptop's IOPS beating a single VM on the first SSD based SAN I deployed as well. Of course, the SAN scaled well beyond it with 1,000 VMs.
I wonder what happened to actual artists though - they seem to be doing fine. I'm sure many people as consumers dabbled in AI art, and reached the conclusion after hours that what they made never looked quite right.
Then they found they could commission an actual artist to draw what they wanted for tens or hundreds of dollars, which is a very good price for getting exactly what you want without having to waste your time playing the token slot machine.
I know some pro artists (ppl doing work for big name companies, games, US film studios), either on a contract or employed basis.
They've always told me the same thing - the job is to hit the minimum acceptable level of quality (which to my untrained eye often looks high, but they reassure me, their work is in fact sloppy garbage), using whatever means necessary, even if that means AI.
They don't even hate AI mostly the way art Twitter does, they hate is because it gives unrealistic expectations to what costs how much, and its often not really possible to get useful results - at least that was the case a couple years ago, things might have evolved.
If AI were good enough, they would certainly use it.
As for Twitter people doing commissions, I dont have firsthand experience, but imo their biggest issue is that there are tons of artists from places like Latam or the Philippines who do high quality work and charge very little, and the people who commission don't care - this was the case well before AI.
As in have the model consider its generated SVG, and gradually refine it, using its knowledge of the relative positions and proportions of the shapes generated, and have it spin for a while, and hopefully the end result will be better than just oneshotting it.
Or maybe going even one step further - most modern models have tool use and image recognition capabilities - what if you have it generate an SVG (or parts/layers of it, as per the model's discretion) and feed it back to itself via image recognition, and then improve on the result.
I think it'd be interesting to see, as for a lot of models, their oneshot capability in coding is not necessarily corellated with their in-harness ability, the latter which really matters.
reply