How expensive is it?
4,912 karma · joined May 21, 2020
DeepSeek v4.1 Flash or GLM 3 are already very very capable models.
Only bottleneck at the moment is the hardware prices.
I'm unable to NOT read the code at all. But I can't read it all either.
I'm more exhausted than before.
Its just that West has been a unified block and no other reason.
Put harsh sanctions on US products and services, few difficult years later there would be a copy of everything US but made in Europe.
But that's a stupid thing to do. West as unified block is the best outcome for everyone but fascist won't understand that.
Nothing much left to innovate.
That's strategic sabotage to say the least.
Ban all open weight models! Cried the VCs.
Basic issue is this:
1. The models are probabilistic in their output. This is objectively verifiable fact. Give same prompt, repeat and you'll get different output.
2. With these markdown heavy approaches, the system has to keep a track of the system state not just in code but in tons of markdown files as well.
3. As the system evolves due to features, bug fixes, the corresponding markdown files are to be updated as well to keep in sync.
4. This update consumes tokens for sure so now token spend goes through roof.
5. But because the models are probabilistic, the real problem is that it is guaranteed that some of the places would not be updated or updated with misinformation in those markdown files.
6. Next agent run when loading the context from these markdown files would have misinformation in its context that it would discover mid task, would have to rectify the situation. token spend would be quadrupled.
7. Only those edge cases of markdown files would hit that has been visited so there would be many inconsistencies still and they would pile up and compound with the passage of time.
8. Anyone who has significant experience with such approaches would testify to that.
9. There's no solution to that other than just using agents as power tools and holding the context and desired outcome in your own head.
Don't they understand e-waste and how is it piling up?
That's why I first switched to Pi and then to DeepSeek harness.
Most LLMs would probably say this piece of code won't compile.
And then using emgui they could expose full fidelity IDE+Agent workflow interface (like DeepSeek Harness and similar) that's also exposed as MCP to be operated by another agent and JSON RPC for over the network but no. Skill issue?
PS: Even full fidelity Photoshop is possible with such high fidelity UI that has CLI, JSON API and MCP server all in the same single binary < 90 MB. For reference, see PhotoCraft or VectorCraft or WordCraft or PdfCraft.
Not to mention it's single binary barely 80 MB and super fast to start and very snappy.
Have you tried that or that PDFCraft?
Basically people are vibe coding their personal apps and anything that's expensive is being vibe coded open in the public. I don't see many software companies staying profitable for long.
DHH is the biggest proponent of AI and let me know which of the 37 signals products can't be vibe coded in a month at a $200 plan that are suitable for that organization alone that just has to be accessible internally only? Hence scale and security aren't such an issue.
In that climate - for how long software companies would stay profitable and when not, who'll be employing developers?
PS: Don't underestimate vibe coded apps. Take a look at PDFCraft, VectorCraft, WordCraft. And imagine the feature parity in a year.
That's writing on the wall.
These are NOT the same thing whereas this one particular suite of clone products is exactly menu by menu, keyboard shortcut by keyboard shortcut and feature by feature a pure clone of corresponding products to that extent that they even have a separate dashboard just to track feature parity: https://github.com/storytold/craft-repo-status-app
It is NOT there yet of course but it won't be there in a year? That's an absurd claim to make. I have checked the repos multiple times and every new release keeps fixing something that wasn't working before.