Tag: AI

Opus

~ Veselin ~ 3 Comments

Claude Opus is my current most favorite model. I had a few blissful months of using it. Generated some good PRs, got stuck in debug loops not as many times as with previous models. I ended up extending the spend limit multiple times.

Opus Cocktail Bar, Sofia

After burning through far too many tokens, I had to stop and think. Is my usage really appropriate? Is it worth thinking how much tokens each prompt consumes? Is it because of the MCPs? Why does it make all these API calls to my dev server? How much does all of that even cost? It’s not clear from the dashboard at all.

While I’m rethinking my life’s choices, I switched to Codex and GPT 5.2. I feel like between the 4 AI editors that I have, I may have enough agent time available to last until the end of the billing period.

Being stuck with one option is not ideal. The situation is not like I have to write code without agents but my overuse of Opus is giving me a glimpse into a future where these models may start costing as much as people.

What Level of AI Use Is Okay for a Blog Post?

~ Veselin ~ 12 Comments

I spend part of my day reading posts and I catch myself developing an allergy to AI slop. At the same time, AI is right into the editor, and attempts to change every word I write incorrectly. So not using it would be foolish. Here’s my gut feeling about what’s fine and what isn’t.

Not Fine

  • Copy/pasting any direct results of a prompt. It deserves a separate post why that’s not fine but for now, it’s a form of bad taste.
  • Any form of GenAI images. They were fun for a little bit. Now, they are just a way to state that the post is AI slop.
  • Let AI change the meaning of content when improving. AI tends to flip the meaning when doing subtle improvements.
  • Engine-oriented titles. If I’m the reader, I want text oriented for humans, not engines. The engines can burn some more CPU and figure it out.
  • Letting AI remove complex words, slang, puns, and emotion. Too much uniformity doesn’t improve readability, it sterilizes the text.
  • Emoji. Thanks to ChatGPT, emoji are like peanuts in a text.

Fine

  • Syntax, clarity, and feedback. AI can help improve the structure and readability the text.
  • Improve individual sentences. I tend to write long sentences and use unconfident words (examples from my post below). Funny that it uses the word unconfident but asks me to remove the word unconfident.
  • Research. Copy/pasting the body of a post to ChatGPT sometimes help find stuff I missed and fact checks. Particularly useful for book reviews.
  • SEO improvements, as long as it doesn’t change the meaning of the content.

Overall, when I catch excessive AI use, I feel an ick about the text. If it feels AI generated entirely, zero chance that I’ll read it.

Why aren’t intelligent people happier

~ Veselin ~ 2 Comments

I found this nice article today that digs into the subject. Check it out.

The article suggests that we’ve been measuring intelligence the wrong way, which leads to poor correlation with life success metrics. Most of our intelligence metrics (like IQ) focus on how well someone can solve clearly defined problems. Real life rarely works that way. Living well, building relationships, raising children, and so on, depend more on the ability to navigate poorly defined problems. As a result, you can have a chess champion who is also a miserable human.

The article goes further and states that AIs can’t become AGIs because they’re only operating with human definitions (training data), and well-defined problems coming from prompts. AGIs would have to master poorly defined problems first.

GPT-5 is out

~ Veselin ~ 2 Comments

I tried it in an 8h coding session. It performs worse than Claude 4 for me, and it’s slower. It made me wait for 10 minutes at a time. Eventually, I gave up, used my brain to understand the problem properly, and hand held Claude to a solution, which took about 1h. I think I lost most of my time with GPT-5 in loops where it fixes one thing at the expense of another while the general approach looked sufficiently sound to fool me but not sufficiently sound to eventually work in all cases.

This might be due to high traffic and not because the model is worse. I’ll give it another chance when the hype fades.