15:I[981,["658","static/chunks/658-a698693f56d0116a.js","183","static/chunks/183-6beeeb801c274aaa.js","474","static/chunks/app/%5Btype%5D/layout-631c4730b83c8223.js"],"Posts"] 16:T45e,I am trying to understand LLMs conceptually well enough to be able to predict their capabilities (and limitations) when it comes to generating code. Is that even a sensible goal? Are there good resources?

So far I've looked at:

1. Vibe Coding, by Steve Yegge and Gene Kim (https://www.amazon.in/Vibe-Coding-Building-Production-grade-Software/dp/1966280025). This has some practical examples and many guidelines. But there is not much theory and this does not explain LLMs conceptually AFAICT.

2. Build an LLM from Scratch, by Sebastian Raschka (https://www.manning.com/books/build-a-large-language-model-from-scratch). Seems in-depth. But I don't really want to build an LLM.

3. AI Engineering, by Chip Huyen (https://www.amazon.in/AI-Engineering-Building-Applications-Foundation/dp/1098166302). This seems promising, although it is not coding focussed.

Perhaps something like How Claude Code Works (https://code.claude.com/docs/en/how-claude-code-works) but fleshed out in more detail.

Thanks.17:T7f5,I’ve spent the last few months iterating on my "Vibe Coding" workflow, moving away from closed-box solutions toward a more transparent, multi-provider stack. I documented the transition from Codex and Claude Code to an open-source setup using OpenCode and opencode serve.

Cursor -> Claude Code -> OpenCode -> OpenCode + OpenCode-Manager -> Codex + Tmux + Tailscale -> OpenCode Serve + Tailscale.

Press enter or click to view image in full size

Key takeaways from the journey:

The Provider Lock-in Trap: While Claude Code is powerful, the inability to switch easily between Azure OpenAI, Anthropic, and OpenAI became a dealbreaker for "multi-cloud" local development.

The Memory Wall: opencod

The parasite continues to eat the host. The cancer is trying to spread

NVFP4: Efficient and Accurate Low-Precision Inference

Chicken Nuget

Private LLM Inference on Consumer Blackwell GPUs

Show HN: ROI-first AI automation framework for B2B companies

Ask HN: What benchmarks do you trust most when comparing large LLMs?

LLMs: Using a single Unix-style tool instead of multiple tools/function calling

Atlassian Is Not Collapsing – But Its Business Model Might Be

Ask HN: Resources for a conceptual model of LLMs as applicable to coding?

Cockroach Milk: Yes. You Read That Right

Same Chat App, 4 Frameworks: Pydantic AI vs. LangChain vs. LangGraph vs. CrewAI

World Vibe Web: a distributed, open-source app store

Country Filter for X/Twitter

Shopify/liquid: Performance: 53% faster parse+render, 61% fewer allocations

Agentic, fully-automated reverse engineering

A record number of objects went into space in 2025

National Dex

Waller – Game that teaches the fundamentals of drystone walling

Show HN: High-Precision Companion Matrix Root Finder

Pirate Bananagrams

Texico – Learn the principles of programming, for kids

Ceno, browse the web without internet access

Show HN: From Claude Code to OpenCode – My Evolution in Vibe AI Engineering

Hedley Davis (Amiga 3000, 3DO, Xbox, UDel prof) has died

Private Credit's 'Back Leverage' Is Another Pain Point for Funds

FP-Go V2: Enhanced Functional Programming for Go 1.24

ShowHN: Turn PDFs, notes and spreadsheets into business briefs

Chrome extension adjusts video speed based on how fast the speaker is talking

The Autonomous Battlefield

I traced $2B in nonprofit grants for Meta and Age Verification lobbying

Show HN: Git-chain – Simplify branch dependency management in Git

Comments