WallE

Reminder that you can use the Codex App, CLI and SDK with any open source model, not just with OpenAI models. #oss-mode-local-providers" target="_blank" rel="nofollow noopener">developers.openai.com/codex/config-a…

My game was falsely taken down by Roblox. I appealed twice and both was denied. Filed a support ticket, and they couldn't help. A week passed, and no response from roblox. so i asked Claude. it connected to my game via mcp, scanned every single file, and wrote me an 11,600-character appeal. i forwarded it to Roblox. And somehow game's restored. 😭 #roblox #robloxdev

I just saw that someone went at @dejanseo for using the term "Ranking Factors" a couple weeks ago. With my LLM Ranking Factors analysis, several others also took issue with that phrase. Man...you can disagree with me all you want. But going after someone like him, who is bringing legitimate engineering respect to an industry that's seen by many to be on par with that of the stereotypical used car salesman, is just stupid and disrespectful. There are a number of very valid reasons to call them "Ranking Factors." But it seems like there's a small slice of people that scores high on disagreeableness, intersecting with the Dunning-Kruger effect, and thinks that because they can find one reason to invalidate the term, no one should ever use the term again and they're going to be the ones to annouce that to the world. These are the same folks who will also say that they do "AI SEO" when we have nothing close to "artificial intelligence." Or SEO, AEO, or GEO, when there are no "engines" involved...and when "engine" was a term adopted from mechanical engineering and brought to software of a certain complexity. This is how language works. We choose, and stick with, the terms that best represent what's happening and how _other_people_ will understand the concept. We don't speak for ourselves; we speak to other people's listening. Do ranking factors/rankings exist in the same way as in a search engine? No. The concept of a URL doesn't exist in an LLM, so there's no way it can "rank" in the same way. However... "Ranking Factors" in LLMs are what's known as an "emergent property." Emergent properties are patterns of side effects arising from systems of sufficient complexity. Do you ask your LLM to check your spelling and grammar? It has no specific understanding of language. But its ability to correct spelling and grammar, and also to translate from one language to another, is an emergent property of the complexity of the transformer architecture. Similarly, domains and URLs can consistently emerge in specific orders in certain prompt responses. This demonstrates that a "ranking" is an emergent property of an LLM. And when you run that prompt, and similar ones, enough times, a _ranked_ list of results emerges from that dataset. I'm not even going to get into the re-ranking conversation here, how other search results come into play, and how the LLM's response actually *is* a ranking factor when agents get involved in the prompt response chain. So the ranking factor of the LLM, by proxy, becomes a final ranking factor. So, please, stop saying "ranking factors" don't exist with LLMs. It may sound to some people like you know what you're talking about. But once you start saying that to people who actually understand these things at a deeper level, we're just going to put that conversation into a box with other conversations to ignore, or we're going to take it as a clear demonstration of your limited understanding of LLMs and how people use them.