Stefano Maffulli

@maffulli.net

Also an architect, GIS enthusiast, sailor.

Research shows that models like DeepSeek-R1-Distill-Qwen-7B can achieve high math accuracy with only 256 thinking tokens, underscoring that longer reasoning is not always better. Convergence signals in internal states indicate routes for more efficient inference. https://arxiv.org/abs/2607.21433

Token Budget Saturation and Mechanistic Early Detection of Reasoning Non-Convergence in Chain-of-Thought Models

ArXiv link for Token Budget Saturation and Mechanistic Early Detection of Reasoning Non-Convergence in Chain-of-Thought Models

arxiv.org

When I was in middle school (late 2000s), I was very interested in mathematics. I thought that the validity of the proof of the 4-color theorem was disputed in mathematics and was shocked when I learned about FLT, but I knew that humans would never beat an AI at chess ever again

Real story: built a GIS database in Access as an architecture student, tracking historical buildings across Andalucía. Worked great — until you needed more than one person touching it at once. That limitation is still the whole story: www.getgrist.com/blog/why-mic... #VBA #OpenSource #MSAccess

Why Microsoft Access survived, and what replaces it | Grist

Microsoft Access survived in regulated orgs because a single person could build a working database without a dev team. What replaces it?

getgrist.com

Telling a permit office to "just rebuild it as a proper web app" isn't help. It's just handing them the same problem with extra steps (dev team, budget, quarters of waiting). I break down why that advice keeps failing: www.getgrist.com/blog/why-mic... #NoCode #VBA

Why Microsoft Access survived, and what replaces it | Grist

Microsoft Access survived in regulated orgs because a single person could build a working database without a dev team. What replaces it?

getgrist.com

Every generation of software produces a new bottleneck for the layer above it. Databases were once the bottleneck until schema documentation and query builders opened them up. APIs were once the bottleneck until GraphQL and OpenAPI opened them up.

How Epoch8 uses Grist to make AI agents query client databases accurately.

When the AI agent gets the query wrong, most teams add more prompt engineering and pray it holds. In this webinar, Olga Tatarinova shows how Epoch8 built the metadata layer instead.

getgrist.com