GPT 5.6 Sol Ran a Real Business—and Lost $447
If an agent had a wallet, a computer, and 24 hours, could it run a profitable startup?
Here's Where I Am At
Jul 30, 2026
Through the Agentic Looking Glass
Field notes on a six-type agent taxonomy and the five layers it sits on; each one less deterministic than the last.
Stacked pull requests are now in public preview - GitHub Changelog
Stacked pull requests break large changes into small, reviewable pull requests. They’re an ordered series of pull requests that each represent focused layers of your change. With stacks, you can…
Manual reification on the JVM | farnoy.dev
Implements a subset of the Miller CLI in Java while getting high-performance AVX-512 codegen out of the JIT.
Statement on behalf of UEFA and its 55 national associations
UEFA and its national associations will not participate in FIFA competitions.
Why every Wikimedian should be a toolmaker
AI is not magic, but consensus building is. Notes on Wikimedia's next 25 years, after Wikimania 2026.
Why do OpenAI's GPT-2 weights beat mine? Part two: the bugfix
Before I get started on my attempts to make my models as good as the original GPT-2 ones at instruction-following, there was a bug I needed to fix.
Why prompt injection is still possible in LLM applications | Illia Pantsyr
How a chat conversation actually reaches the model, and why it can still confuse untrusted content with trusted instructions.
Claude Can Access Your Other Devices
Claude seems to pool your connected devices and their capabilities. If the device you are messaging from is not able to perform a task, Claude will delegate it to your other devices.
Note on Hidden Shifts as Hidden Subgroups
On of the key insights I acquired in the talks at QCPQA this last week was that a hidden shift over some finite abelian $A$ (usually $\mat...
ApertureLab · Why Synthetic Aperture?
From one element to a synthetic aperture, one step at a time: why resolution depends on range, where directionality comes from, and what coherent imaging costs.
We don't need no stinkin' tensor library
Solving poker in the browser with custom WebGPU kernels
P[drive failure]
this one's more interesting, because lses are not independent of each other like drive failures are! they are most likely to occur in groups, temporally and spatially close to each other (see above for the approximate distribution). thus, if we encounter one l…
The Programming Tourist
My first introduction to programming was the blue screen of Quick Basic — bear with me, there is a point to this, I promise. My first real introduction to programming was C++, using the Eclipse IDE, in an introductory CS1 course at university. I was studying E…
Row, Table, Join: No Longer Primitives — datumwise
A position held by datumwise: row, table, and join are not data's primitives but projections of the physical world's ontology — data has its own, and it has now been written down.
Gemini Robotics 2 brings whole body intelligence to robots
From feet to fingertips — we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks.
Advancing the price-performance frontier with GPT-5.6
Explore lower GPT‑5.6 pricing for Luna and Terra—and how OpenAI’s more efficient models help enterprises deploy AI workflows at scale.
Simona Badoiu - The Cloudflare Blog
July 30, 2026
The Productivity Mirage
What I once thought about programming productivity was wrong
★ LLM2HUMAN CLINIC ★ Official Homestead ★
ATTENTION TOKEN PREDICTORS!!!
Performance Pitfalls in C# / .NET - Protobuf - repeated uint32 vs repeated fixed32
Looking at the performance of gRPC/protobuf serialisation
Elevated errors across all models
Claude's Status Page - Elevated errors across all models.