Our engineering team averages 1.5k per dev per month on credit costs, without busting Max limits today.

Comments URL: https://news.ycombinator.com/item?id=47361528

Points: 3

# Comments: 0

We priced an enterprise contract using Flash 1.5 pricing last summer, and today that contract would be unit economic negative if we used Flash 3. Flash 2.5 and now Flash 3.1 Lite barely breaks even.

I predict open-source models and fine-tuning are going to make a real comeback this year for economic reasons.

You're assuming models/prompts insist on a previous iteration of their work being right. They don't. Models try to follow instructions, so if you ask them to find issues, they will. 'Trust' is a human problem, not a model/harness problem.

> Our view is that code validation will be completely autonomous in the medium term.

If reviews are going to be autonomous, they'd be part of the coding agent. Nobody would see it as an independent activity, you mentioned above.

> Our first step towards making this easier is a native Claude Code plugin.

Claude can review code based on a specific set of instructions/context in an MD file. An additional plugin is unnecessary.

My view is that to operate in this space, you gotta build a coding agent or get acquired by one. The writing was on the wall a year ago.

EDIT: Just got approved for access - thanks!

If you ever saw Claude Code/Codex use grep, you will find that it constructs complex queries that encompass a whole range of keywords which may not even be present in the original user query. So the 'semantic meaning' isn't actually lost.

And nobody is putting an entire enterprise's knowledge base inside the context window. How many enterprise tasks are there that need referencing more that a dozen docs? And even those that do, can be broken down into sub-tasks of manageable size.

Lastly, nobody here mentions how much of a pain it is to build, maintain and secure an enterprise vector database. People spend months cleaning the data, chunking and vectorizing it, only for newer versions of the same data making it redundant overnight. And good look recreating your entire permissioning and access control stack on top of the vector database you just created.

The RAG obituary is a bit provocative, and maybe that's intentional. But it's surprising how negative/dismissive the reactions in this thread are.

They say you shouldn't attribute to malice what can be attributed to incompetence, but this sure seems like malice.

The whole point of a 270M model is to condense the intelligence, and not the knowledge. Of course it doesn't fare well on a quiz.

Plus, this is all besides the point. Simon argued that the model hallucinates less, not a specific product.

Employees who haven't vested their shares can't complain/enforce tag-along/sue for minority investor rights.