Every issue, in one place

The Everyday Brief #10: a CTE that returned $680 instead of $915, frontier models falling to chance at 10,000 tokens, and a $500 STEM scholarship closing December 18

Five CTE traps, each measured on twelve rows: a filter in the wrong place moved a total by $235. StateBench shows frontier models fall to chance on state tracking at 10,000 tokens. Sunflower V2 covers 67 African languages, and the WAAW Foundation scholarship for women in STEM closes December 18.

November 10, 2026

The Everyday Brief #9: a phishing test that passed until the attack adapted, a plan check that cut a lookup from 5.09ms to 0.02ms, and Gates AI Fellows closing November 11

Six email agents caught every old phishing email, then an adapted attack lifted Qwen susceptibility from 0.12 to 0.95. One index cut a point lookup from 5.09ms to 0.02ms. East Africa approved a shared AI module for undergraduates, with adoption still voluntary. And Gates Foundation AI Fellows close November 11.

November 3, 2026

The Everyday Brief #8: a mean shift that left AUC untouched, a harness swap that was mostly noise, and Girl Up awards closing October 31

A mean shift left AUC exactly unchanged at 0.8834 while accuracy fell from 0.80 to 0.60. Swapping agent harnesses flipped 13% of tasks, the same as rerunning the same harness. Three plans to share AI computing across Africa landed in six days. And Girl Up Project Awards of up to $1,000 close October 31.

October 27, 2026

The Everyday Brief #7: five leaks that inflated AUC by up to 0.26, a true tool result that was still wrong, and $10,000 awards closing November 13

Five controlled experiments put a price on data leakage: target leakage bought 0.16 AUC, feature selection before the split bought 0.26. ReplayBench replayed real tool results into the wrong task and agents acted on them in 64.9% of cases on average. Togo is paying people to record Ewe, Kabiye and Moba. And three Lacuna Fund awards of up to $10,000 each close November 13.

October 20, 2026

The Everyday Brief #5: a harness that doubled an agent's score, five SQL patterns that beat self-joins, and a fellowship closing October 15

A visual-memory harness took Claude Opus from 40.68 to 100.00 on ARC-AGI-3 without a smarter model. Five window-function patterns cover most analytics SQL. Google backed 15 revenue-generating South African AI startups. And a fellowship auditing AI products in local context closes October 15.

October 6, 2026

The Everyday Brief #4: the frontier price war, an honest benchmark, and a compute grant closing tomorrow

Frontier API prices fell 40 to 60% in 48 hours. DuckDB swept our 3.5M-row benchmark. A Nigerian startup rents GPUs under $1 an hour. And a compute grant for African AI closes tomorrow.

September 29, 2026

Google's AI tutor worked. It worked best for the kids who needed it least.

Plus: the 20% of SQL that does 80% of the work, and a funded AI lab in Accra with a deadline.

July 14, 2026

Welcome to The Everyday Brief

Practical AI, ML & data science, every week, here's your first issue.

July 12, 2026

The 20% of SQL that does 80% of the work

Plus: AI agents that run the whole data-science pipeline, Africa's first applied AI lab opens in Accra, and a funded accelerator with a deadline.

July 12, 2026