15:I[981,["658","static/chunks/658-a698693f56d0116a.js","183","static/chunks/183-6beeeb801c274aaa.js","474","static/chunks/app/%5Btype%5D/layout-e8e83705b5c8d3b7.js"],"Posts"] 16:T6a3,I'm just a hobbyist that has ran LLM models locally and follow a lot of content about it. Hope we have a few AI researchers here on HN to clarify this.

When using Opus or Codex vs. a chinese or Open source model, it feels like its reasoning capabilities are basically the same.

The difference is typically in coding. It looks like OpenAI and Anthropic invest a lot in pre-training (paying Mercor and the like).

Also a lot in creating synthetic data, I believe this has bigger AI research involvement and techniques.

This ends up creating the perspective that it is smart, after all, it has been trained with what you want to do, so it can do that for you.

But overall, is there really much AI research being done on those companies, or are the AI researchers mostly fine-tuning small aspects of the model, akin to what Google engineers used to do for Google search?

Of course, there's the RLHF loop that developers using Anthropic/OpenAI products as well, which provides probably very good data.

I ask this because this all looks like somebody with money could throw money at the problem and end up with a better model at the end, provided they do what I outlined above better -- with AI research being really not that important.

It still often feels like talking with ChatGPT 4 with just better data.

Even the big upgrade of Claude Code being able to work autonomously looks to be mainly due to it knowing how to grab context and do tool calls (not saying that this is easy), rather than the model's raw performance being better.

Or am I wrong, is there something extremely good on those models that AI researchers discovered that the others don't have?

Show HN: We made 3M jobs searchable on a globe to visualize labor markets

MacPulse – macOS System Monitor and Performance Logging

Show HN: Protovoters -- Free, accessible voter files for democracy

Le Glaude Code – A CLI that responds to Git command with quotes from a Fr movie

Show HN: Desktop OSD that reads Claude Code's rate limit headers

A robot is beating human pros at table tennis

Anthropic is A/B testing removing Claude Code from Pro plans

TalentRecruit

Corral: Measuring how LLM-based AI scientists reason, not just what they produce

FreeBSD Status Report First Quarter 2026

Texas launches $750M plant, targets growing New World screwworm threat

Ask HN: Is the ongoing AI research driving LLM models to be better?

The Prevention of Literature (George Orwell, 1946)

Microsoft issues emergency update for macOS and Linux ASP.NET threat

I blind A/B tested 40 Claude prompt codes, only 7 shift reasoning

Stop Hardcoding Your Timeouts

Retractions of scientific papers by Nobel Prize Winners

Notes on writing a voxel game in Dyalog APL

Signaling Is the Intelligence

The Log Is the Database

Ask HN: Linguistics/Computing Overview for the Layperson

Production MCP Patterns, Part 2: The Agent Stack Grew Up

Light Pollution and Wildlife Interactive Ecology Simulation

Reverse-Engineering a North-Korean-Style Supply Chain Attack

I was reported by ChatGPT officials

Augmenting Human Intellect (1962)

US health officials Nix publication of a study on Covid vaccine effectiveness

Show HN: Vibeyard – An open-source IDE for managing AI coding agents

Honey, I Shrunk the Coding Agent

Software Eats Its Own

Show HN: Linux Desktops in the Browser