honestly its gotten more mixed. diffusion in text, AR in images. hybrid archs everywhere (altho most researchers tell me the differences between SSM vs transformers not that material). not clear.
@swyx
-
MRL Embeddings: 10x Size Reduction with Minimal Performance Cost
By
–
remembered that MRL embeddings exist and immediately reduced my embedding size from 6MB to 600KB for a 5 point drop in MTEB lol
-
Google DeepMind API Key Enables Full AI Development Freedom
By
–
just realized the nice thing about having @GoogleDeepMind as presenting sponsor is i can literally do every damn thing i want to do in AI with one api key https://t.co/FLab1FPh40 pic.twitter.com/lxSPgbRNhj
— swyx 🐣 (@swyx) 16 mars 2026just realized the nice thing about having @GoogleDeepMind as presenting sponsor is i can literally do every damn thing i want to do in AI with one api key
-
Cognition’s Growth Formula: Cloud Computing Multiplied by Coding Agents
By
–
btw the @cognition growth formula is very simple: Devin growth = cloud komputer * koding agents growth math it out. working on releasing some numbers soon
-
Starting AGI-Focused Track at AI Engineer World’s Fair
By
–
i think this is enough motivation for me to start a "how to be more AGI pilled" track at https://
ai.engineer/worldsfair . usually we're more narrow/specific/practical, but i think this is an important enough recurring theme that people should just generally be practicing. -
AI Automation Limits: Testing Claude Coworking Chains and Judgment
By
–
my version of 2 is “the stuff that AI just oneshotted, what was the immediate action or thought just before or after you handed over to ai” – for me specifically claude cowork. i end up chaining more and more things and testing the limits of the automation and judgment and
-
Developing AGI-Awareness as a Learnable Skill
By
–
> 2) Being AGI-pilled is a skill & you can get better at it can you give more detail/thoughts now?
-
Nemotron’s Intentional Limitations in AI Model Design Strategy
By
–
i have a pet theory that Bryan and co cannot make Nemotron TOO good bc otherwise it would scare off customers. it needs to be “good enough to reflect everyone’s baseline consensus on SOTA techniques to lower the bar for neolabs + be maximally useful for synthetic data bc that is
