Google today is not the same company today as it was yesterday

Is DNS hard, or does GoDaddy just really suck?

Yo, DNS is harder that inventing new computer science

LLMs in coding harnesses are 100% neurosymbolic programming. One may ask: Is NSP doomed to this arrangement, with tools and programs bolted on after-the-fact? If you disagree like I do, check out this program synthesis experiment, Kelp: github.com/Open-Athena/.... What is the right middle ground?

GitHub - Open-Athena/Kelp

Contribute to Open-Athena/Kelp development by creating an account on GitHub.

github.com

Programming is built on paradoxes. Reality has an infinite amount of detail. The programs will get more abstract, powerful, expressive, etc., but they will never go away.

𝓐𝓾𝓫𝓻𝓮𝔂@aub.bsky.social · 3d ago

maybe we get higher-level high level programming languages out of all this, for exactly that reason 🤔

A program is a specification for what a computer should do. By this definition, “vibe coding” is also coding. It matters not whether the specification is in assembly, Python, or English.

Hell yeah! Banger (gift) article www.nytimes.com/2026/08/02/o...

Opinion | The ‘Civil War’ Is a Lie. Let’s Call It the ‘Slaveholders’ Rebellion’ Again. (Gift Article)

To face our own history, we must be honest about it.

nytimes.com

You will live to see your greatest intellectual aspirations dismissed as trivial

Here's the tale of how @jder.bsky.social and I scaled Samudra, a neural ocean emulator capable of predicting 8 years of the ocean on a single GPU, to operate at a full 1/4° resolution (16x the size in bytes). It was quite a humbling process.

Open Athena@openathena.ai · 6d ago

Simulating ocean climate takes a supercomputer 4,600+ CPU cores to produce 12 simulated years per day (SYPD). Samudra 2 produces 4,800 SYPD on 1 GPU at the same resolution. In a new blog, @al.merose.com reports on Samudra, a neural ocean emulator built in collaboration with NYU & MIT: bit.ly/oa-ss

I am delighted to announce that after 5 years of work, 15 early releases, 50,000 words written, and 100,000 more deleted, *Logic for Programmers* is now complete and available in print! Blog post announcement: www.hillelwayne.com/post/lfp/ Website: logicforprogrammers.com

Logic for Programmers

Practical math for the working programmer.

logicforprogrammers.com

honestly those people don't know anything

The trillionaire owner of Tesla and SpaceX and a major media org is posting about how we need to expel all non-white people from their countries and kill the people who oppose this. And it’s just…fine? People will still show up to work for him?

I wonder if latex support in the Bluesky client would get all the researchers over here.

I find myself writing the initial design doc with Opus 4.8, then performing adversarial refinement with Fable. This seems reasonably toke efficient to produce a high quality design.

The AI community failed to exit twitter. What implications can we draw about what people believe from this?

What it means that the AI community can't quit twitter

Some quickly jotted thoughts working through the implications of the AI community remaining on twitter

rl-blogging.leaflet.pub

I appreciate DHH's thoughts, but I disagree on where Europe's problems are, and from where I'm sitting it's not immigration. Thus I wrote a response to his post. dark.ronacher.eu/2026/7/17/li...

Europe Won’t Live By Deporting

A case for focusing on Europe’s actual problems.

dark.ronacher.eu

Claude asked for approval for calling `rm` on a local log file (metadata, meaningless), but did not ask for approval and went ahead and ran `gsutil -m rm` on a data corpus used for training 🙃

We’re on the front page of HN

Al Merose@al.merose.com · 4w ago

Here is my talk: “Geospatial ops are relational.” I make some bold claims here: - Specialized climate and geo libraries are unnecessary. You just need the right data model and SQL - Differentiation might be the first new component for RDBMS in decades - ML models are differential databases

New twiggy pic just dropped

Here is my talk: “Geospatial ops are relational.” I make some bold claims here: - Specialized climate and geo libraries are unnecessary. You just need the right data model and SQL - Differentiation might be the first new component for RDBMS in decades - ML models are differential databases

DataFusion Community Showcase Vol. 1: Whereobots/SedonaDB & Xarray-SQL

YouTube video by Spice AI

youtu.be

Al Merose@al.merose.com · 4w ago

DataFusion showcase went great! Here are my slides: docs.google.com/presentation...

webgpu wind line lod stuff

I’m too young to be in the crackpot phase of a research career

DataFusion showcase went great! Here are my slides: docs.google.com/presentation...

Geospatial Operations are Relational Operations

Geospatial ops are relational: An introduction to Xarray-SQL Alexander Merose | July 2026 alex.merose.com @al.merose.com

docs.google.com

Al Merose@al.merose.com · last mo.

Sneak peek: I implemented a 3 layer MLP that predicts the fashion MNIST dataset using SQL (with my own autograd impl inside the database). Apparently einsum notation is better expressed as a JOIN. github.com/xqlsystems/x...

Betsy, my co-worker at OA, is doing inspiring work on materials science (see tomat.oa.dev). Check out this behind the scenes blog post to learn more about her story. 🏺

Open Athena@openathena.ai · 4w ago

Betsy Cannon is a member of technical staff, where she leads material science projects RHOAR-Net, with Princeton's Rosen Research Group, and MarinMat. Read her story on learning DFT, how AI changes shared research infrastructure, and what pottery taught her about building by hand: bit.ly/oa-betsy

I'd like to announce that at @openathena.ai, @jder.bsky.social and I helped @m2lines.bsky.social release Samudra 2. We scaled this neural ocean emulator to train on 16x the size of data in bytes on the same hardware budget. We can now skillfully predict 8 years of the ocean on a single GPU at a 1/4°

🌊 Samudra 2: A Fast, Cheap AI Ocean Model, Now at the Scale That Matters

M²LInES’ neural ocean emulator now runs multi-year simulations at eddy-permitting resolution on a single GPU, turning a supercomputer-scale…

medium.com

If you’re a true believer like us (SQL is better than Python for ML), consider joining my laboratory discord, link at the bottom of xql.systems

stellz@piss.beauty · 4w ago

%100 serious that I think sql would be a much better way to express neural networks than python

Yes, indeed, this is cursed. This is the ghost of Michael Stonebraker’ SciDB, which is also cursed. www.theregister.com/software/201...

SciDB: Relational daddy answers Google, Hadoop, NoSQL

Stonebraker doesn't drop ACID

theregister.com

Cameron@cameron.stream · 4w ago

Al Merose@al.merose.com · last mo.

Sneak peek: I implemented a 3 layer MLP that predicts the fashion MNIST dataset using SQL (with my own autograd impl inside the database). Apparently einsum notation is better expressed as a JOIN. github.com/xqlsystems/x...

Sneak peek: I implemented a 3 layer MLP that predicts the fashion MNIST dataset using SQL (with my own autograd impl inside the database). Apparently einsum notation is better expressed as a JOIN. github.com/xqlsystems/x...

xarray-sql/benchmarks/nn.py at claude/xarray-sql-mnist-demo · xqlsystems/xarray-sql

An experiment to query Xarray datasets with SQL. Contribute to xqlsystems/xarray-sql development by creating an account on GitHub.

github.com

Al Merose@al.merose.com · last mo.

It turns out SQL is an even simpler for writing NNs than JAX. With this arrays-are-pivoted-tables notion, then vmap completely goes away. You just have `grad’, `jvp` and `vjp` on columns which are Nd arrays. “Hard” tensor math is a dumb JOIN. matmul is a JOIN and GROUP BY and SUM.