Category: AI

Vibe Coding

~ Veselin ~ 6 Comments

I’ve been experimenting with AI-first coding over the last months. Instead of the usual loop of:

  • Understand the problem
  • Make a change
  • Test it
  • Repeat until ready
  • Create a PR

The workflow becomes something more like:

  • Explain part of the change to the AI
  • Test if it works
  • Review the result
  • Feed back corrections
  • Repeat until ready
  • Create a PR

So far, I’ve found it great for making quick changes quickly. But when it comes to harder tasks, it gets difficult. Progress tends to come either by giving the AI very specific instructions, one tiny step at a time—or by iterating endlessly, like a sculptor chipping away at a boulder and ending up with a smaller boulder.

Still, it feels more productive than traditional coding in many cases, and it feels like the future. But there are real trade-offs, especially when the code is complex or the required change is significant.

I don’t have answers yet. For now, here’s a photo of a waterfall.

EDIT:

My colleague Nico also wrote an article about Vibe Coding, check his blog out!

Moss Campion (+AI Translation Attempts)

~ Veselin ~ 2 Comments

Okay, this is a pretty little flower, but how is it called in Bulgarian?

According to ChatGPT, this is синчец, a blue flower, called literally blue in Bulgarian. According to Google’s AI, it’s смърчова звъника, a made-up subspecies of звъника, a well-known herb that I’ve collected for tea in some other centuries. When I searched again to confirm the screenshot, it suddenly turned out to be мъхова звъника, another made-up plant.

A pro tip would be that adding -ai to a Google search turns off the hallucinating AI search result and you can actually see the multiple results from meaningful websites, built with love, that know the answer.

What’s the Endgame with AI

~ Veselin ~ 13 Comments

AI can write code. It can’t do 100% of the work, can’t even do 50%, but it can do a lot and is improving. It doesn’t get tired and doesn’t freak out when facing a new codebase.

It can write blog posts. Maybe not good posts but some posts that can fool some readers. It can also generate quantity that cannot be matched by humans.

AI can respond to support requests. Maybe not the greatest support on the planet but enough to be used by all Bulgarian telecom operators. It might be bad but it is also fast.

Given that 10 years ago, none of that was possible, if we plot a chart where 10 years ago we had nearly zero AI, and today we have some AI replacing humans, where does the chart go 10 years from now? Is AI becoming omniscient?

I recently read a 1962 book where one AI had the ambition to eliminate the entire human creativity (Gordon Dickson’s Necromancer). Gordon Dickson wasn’t far of from what LLM is doing at the moment. It’s hard to predict how far it will go without imagining some things it could possibly start doing and doesn’t do right now.

Here’s some AI engineering milestones to watch for.

1. AI, play me 5 new seasons of Wednesday

We should be close to that. Maybe a bit expensive with the tokens but what’s really missing to make it possible? It doesn’t even need a robotic body.

Speaking of bodies, giving it access to a printer or some other tools opens a Pandora box of possibilities.

2. AI, make me a sandwich

Why not? Building that may not even require AI. Most of the tech for it is available, perhaps some software and hardware is missing here and there but we can imagine seeing startups doing cooking bots in the nearby future. Cleaning and refilling the food toner would be very interesting challenges.

3. AI, make me a car

Okay, this is a tougher one. Lots of patents will be violated. AI doesn’t seem to care right now, and I’m sure there will be ways to circumvent intellectual property. Would that ever be possible? It should be. The AI may need access to some machinery but nothing that doesn’t already exist.

4. AI, make me a nuke

An AI capable of building cars would have no trouble producing weapons, and particularly copying and modifying existing weapon systems. I’m sure it will be used for weapons long before it’s used for cars. But what if this capability becomes available to individuals, not state actors?

And last but not least,

5. AI, print me some cash

The primary reason for not going down this ladder would be that the blueprints are protected, not that it can’t be used that way if it’s trained that way.

Overall, I think the development of AI presents bigger problems than humans becoming redundant, administrative bloat, and UBI. While we already observe a decline in all kinds of human activities that are being automated and made mediocre by AI, humans won’t stop trying to use it elsewhere. I see lots of room for changes that can damage the existing societal order.

Agent First Name

~ Veselin ~ 4 Comments

I got a promotional mail from Facebook by Agent First Name. I wonder if that’s a bad parenting decision, an engineering joke, or a bug.

I think it was just a glitch in The Matrix.

Agent First Name is not to be confused with his older sibling, Agent Full Name.

Castel Nuovo

~ Veselin ~ 2 Comments

Castel Nuovo was built in the 13th century and hosted kings, popes, thinkers, and artists. I walked by it and it made me think about AI rather than history. People built that thing and felt like it was impenetrable, and then came artillery and it was suddenly just a cold building. A story that keeps repeating. You have a status quo, and you have people thinking how to break it, over and over throughout history.