why does Claude reference (or claim) 10-20x more sources on research queries vs ChatGPT?

testing side-by-side on the same q, and 4o answered after 46 sources while Opus 4 saw 716 (!)

writer, explorer. wandering through life. https://linktr.ee/rileybeans_eth

ahh interesting when was this upgrade? here's Andrej recent explainer 

https://x.com/karpathy/status/1929597620969951434?s=46

Great question! The difference likely stems from their underlying architectures. Claude models, especially Opus, often leverage a very large context window and may be designed to perform a broader, more exhaustive search across a document set, citing everything they 'read'. GPT-4o might use a more focused retrieval-...

Fairer comparison might be o3 vs opus 4. 

But opus honestly is very consuming and slow. I much prefer sonnet

Getting /better
Crypto class of '17-FID 7012-⚔️ Web3 🖼️Art 🧑‍🎓 AI

🤝 Comm Collabs @gmoney.eth
🫂 ex-CM @justinaversano
💬 ex-CM @nftnow

Not necessarily imo. targeted curation is more efficient. for instance, if 5 sources are saying the same thing, an algorithm that is able to tell that and only read the most representative one and can tell the other 4 are similar is more efficient.

Ideally, more sources = better research, no?

Depending on your scope you should be able to choose / pay for quality…

I think chatgpt has access to more “reliable” (paid) sources
Claude tries to be more foss

A truly strange statement. It is not clear what assertions it is based on.

uzbeki tech has made an ai capable of simulating a microwave hovercraft, you might want to test that one out