Jesus this is inspiring.

RAG Is Read-Only Memory Andrej Karpathy put a name to the architectural flaw breaking enterprise AI: RAG treats retrieval as understanding and re-derives knowledge from scratch on every query. The missing piece is a compilat...

I _really_ like what Karpathy is talking about with LLMs and flat file memory. What do I do when i like something? Write a blog about it of course.

www.distributedthoughts.org/2026-04-20-r...

*ahem* usemakoto.dev

Can't respond to this but.. YES. SOURCE MATERIAL NEEDS MANIFESTS.

Blargh thank you!

shit man, i could only stay secret for so long

The ASIC Unbundling Inference is two-thirds of all AI compute. Almost all of it still runs on chips designed for training. The market is finally noticing the mismatch.

you: "hey dave! when are you going to talk about ANYTHING else besides edge compute?"
me: "NOT TODAY MOTHERFUCKER"

www.distributedthoughts.org/2026-04-16-t...

As everyone knows, the Catholics have famously never had an army or owner 2/3 of Europe

yeah, that was the inspiration for the post!

i will not accept this kindness without offering to buy you a case of chianti.

Agents Don't Live in Data Centers Agentic AI is the first workload where the physics of latency and data locality aren't optional. The $700B centralized infrastructure bet was built for a different kind of AI.

AGENTS! (don't live where you want them to live)

www.distributedthoughts.org/agents-dont-...

Chef Reactions on Instagram: "Off pudding @elainecarolskitchen" 26K likes, 4,250 comments - chefreactions on April 10, 2026: "Off pudding @elainecarolskitchen".

I watch a lot of food content. This may be the worst I’ve ever seen. Not an exaggeration. www.instagram.com/reel/DW9As21...

10B cameras by 2030. 99% of that video is never analyzed.

So we ran @GoogleAI new Gemma 4 model directly on edge hardware - Object detection. Scene description. Text reading. Safety scoring. All on-device. Zero cloud.

Full demo: exso.cloud/google-gemma...
Github Repo: exso.cloud/google-gemma...

I’m as fucking surprised as you!!

The Grid Said No Eleven gigawatts of announced data center capacity sits frozen in 2026 without construction underway. Not because of chip shortages. Not because of capital constraints. Because the electrical grid cannot deliver the power. That number deserves a moment. Eleven gigawatts is roughly the output of eleven nuclear power plants. It's enough

as always, you just can't techbro your way around physics. :(

www.distributedthoughts.org/2026-04-09-t...

Man…. It’s not the apocalypse, but so many group chats I’m in rn

Nfw

The Inference Inversion For the past three years, the AI industry has operated under a simple assumption: more centralized compute solves everything. Bigger clusters. Bigger data centers. Bigger power contracts. The logic wa...

Where i talk about (what else) inference at the edge

www.distributedthoughts.org/2026-04-06-t...

i'm sure their servers are meltiing, but this move by Anthropic could be... not good. Driving your users into the hands of your competitors is tough.

I think it takes a special kind to see the green and white logo, the bookstore theme, the name of the store, and the time in which it was made to think it was a 1:1 retelling of a play in 1937 with no relevance to the current state. But you do you!

"Imbattables" - France World Cup 2026 Song YouTube video by All About Barca

Dude this shit is HARRRRRRRRDDDD

youtu.be/Ym5tRqf37eI?...

It’s wild that they made an entire movie about how evil Barnes & Noble (You Got Mail) was and now they are the plucky underdog that everyone roots for over Amazon

Justice for Casey

One Year of Liberation Day: What the Tariff Rollout Actually Revealed About AI Infrastructure One year ago today, President Trump stood in the Rose Garden and imposed double-digit tariffs on virtually every product the United States imports. He called it "Liberation Day." Within weeks, tariff ...

Today's the 1yr anniversary of Liberation Day tariffs. The AI buildout needed a $34B/month emergency exemption to survive. Maybe don't build all your infrastructure where one policy change breaks everything?

www.distributedthoughts.org/2026-04-02-o...

[No Action, Just Support!]: From the Co-founder of Kubeflow · Issue #484 · NVIDIA/aicr Hey AICR team! I wanted to drop a note here because this project genuinely made my day. I was very very early on Kubernetes at Google, and then co-founded Kubeflow. The whole dream from day one was...

I am REALLY excited about the AICR project from NVidia and others.

github.com/NVIDIA/aicr

This was exactly the direction I wanted Kubeflow to go, and I'm very happy someone has picked up and run with it!

github.com/NVIDIA/aicr/...

technically correct is the best kind of correct

what do the attorney general and an iMac G3 have in common?