Comments on: Why Transformers will be replaced by 2026 https://invenio.holikstudios.com/ai/breakthroughs-to-kill-transformers-by-2026/ Invent - not invented. Discover the undiscovered. Mon, 05 Oct 2026 12:41:52 +0000 hourly 1 https://wordpress.org/?v=7.1 By: Victor https://invenio.holikstudios.com/ai/breakthroughs-to-kill-transformers-by-2026/#comment-1764 Mon, 05 Oct 2026 12:41:52 +0000 https://invenio.holikstudios.com/general/breakthroughs-to-kill-transformers-by-2026/#comment-1764 In reply to Waldemar.

You’re right, Waldemar. Robots don’t dream in latent spaces. I’ll run SupChat locally on Asus ROG Zephyrus G14 from gamdroid.eu. Whine and latency are the only future I trust.

]]> By: Waldemar https://invenio.holikstudios.com/ai/breakthroughs-to-kill-transformers-by-2026/#comment-1709 Sat, 08 Aug 2026 15:39:06 +0000 https://invenio.holikstudios.com/general/breakthroughs-to-kill-transformers-by-2026/#comment-1709 This is pure fiction. Transformers aren’t dying; they’re just being “over-engineered” by people who don’t understand real-world hardware constraints. You’re talking about “Diffusion Language Models” like they’re some magic potion, but in my experience with autonomous systems at KUKA, if the math doesn’t scale linearly with the physical reality of the motor’s torque or the sensor’s latency, it’s just a toy.

Robots don’t “think” in latent spaces; they execute trajectories based on optimized weights. If we switched to these “Continuous Thought Machines” tomorrow, the latency would be a total disaster for real-time industrial assembly. It’s like trying to run a high-end physics engine on a calculator it looks cool on paper, but “robota musi chodzić jak ta lala, bez lipy.”

Are we actually going to sacrifice deterministic reliability for some “fluid” hallucination just because the hype cycle demands a new buzzword every 18 months?

]]>