Sorry, but this is (mostly) wrong. Do they *need* XML? No. But does it improve outputs dramatically for complex tasks? Hell yes.
LLMS
-
Markdown Performance Limitations with Claude AI Models
By
–
Agree, though I've found Markdown performs slightly worse on my tasks, esp. if using Claude
-
XML Prompting Outperforms JSON for AI Models
By
–
Of course, you can get way more in-the-weeds with this (see some of my prompts on my timeline), but truly, basic XML prompting is this simple. It'll almost always outperform JSON prompting, is easier to write, and easier to read.
-
Reference to divergent behavior in AI models
By
–
Looks similar to the divergent behavior elicited in the “repeat poem forever” paper:
-
Brain Vision Processing Gap in Modern LLMs Architecture
By
–
the human brain reserves 40% of its processing exclusively for vision. modern LLMs somehow evolved without this entirely
-
Turing Award 2025 honors attention and transformer inventors
By
–
The 2025 Turing Award will go to Dima Bahdanau and Noam Shazeer for inventing attention and transformers.
-
VLMs Outperform with Visual Data Alone Over Numerical
By
–
Interestingly, VLMs perform better with plot alone compared to plot + data! This paper shows that GPT‑4.1 and Claude 3.5 excels at scatterplot analysis, and could often perform better only by looking at the plot, without reading the actual numerical data.
-
Gemini 2.5 Conversational Segmentation: Natural Language Vision AI
By
–
🖼️ Conversational Segmentation (Gemini 2.5): hype or helpful?
— Louis-François Bouchard 🎥🤖 (@Whats_AI) 28 juillet 2025
• Segments with natural‑language prompts (vs Meta SAM’s points/clicks/boxes/masks).
• Handles relationships, conditionals, OCR text, and multilingual labels.
• Great for prototyping and LLM apps; edge/real‑time… pic.twitter.com/APadBDIMn6Conversational Segmentation (Gemini 2.5): hype or helpful? • Segments with natural‑language prompts (vs Meta SAM’s points/clicks/boxes/masks).
• Handles relationships, conditionals, OCR text, and multilingual labels.
• Great for prototyping and LLM apps; edge/real‑time