Paul Watson 🌍

@paulmwatson.com

Web developer born in the last millennia. Vice-chair somersetstrikers.co.za. South African. 1/64th Irish.

Company: Claude, write a job spec for this role Applicant: Claude, rewrite my resume for this job spec Company: Claude, analyse resume for this job spec ... Company: Claude, hire applicant ... Applicant: Claude, do the job

Tapper, Tapper, stop, listen, Tapper, ok, no, Tapper, will, Tapper, you, Tapper, stop, Tapper, you’re making me, Tapper, angry, Tapper, Tapper, why are you, Tapper, oh, Tapper, stop it, Tapper, Tapper, *twitch*

Shieldstral: you write the policy as a plain-language question at inference time, and the model returns a calibrated safety score. No retraining, one interface for text and images, and a verdict from a single token. Colour me skeptical.

Sung Kim@sungkim.bsky.social · 24h ago

Mistral AI's Shieldstral (open-weight) A 3B open-weights, policy-adaptive multimodal safety classifier that matches models up to 7x its size on text safety and sets a new state of the art on multimodal moderation. mistral.ai/news/shields...

The variance is larger, too. From 1895-1915, the maximum difference from the rolling average was +/-10. In the last 20 years, the difference ranged from -10/+15 and the negatives (cooler than average) are diminishing.

I must say this is one of the best things I've read in days. Absolutely dripping with charts and insights on the scale and intricacies of the wildfire season in Europe, as things currently stand. It is a complex picture - and yh, it's *bad* in many places. hannahritchie.substack.com/p/europe-wil...

Is Europe having a bad wildfire year?

Several countries in Europe are having a very large wildfire year; this is offset by lower burn in the Balkans

hannahritchie.substack.com

No change in model accuracy isn’t something you can guarantee. Quantization is a lossy compression technique, so information is discarded. In practice, providers usually mean benchmark scores show no statistically significant drop, not that every real-world inference will produce identical outputs.

Cloudflare@cloudflare.social · 2d ago

Serving frontier models like Kimi and GLM means fighting for GPU memory. Here's how we quantize KV caches, compress model weights, and add integrity checks to serve them faster, cheaper, and safely. https://cfl.re/4bWZltY

Same as it ever was: The fundamentals of getting your content seen in the age of AI are clear writing, strong structure, and trust signals. Take our free classes to learn what _has_ changed and what you can do about it.

WordPress.com@wordpress.com · 6d ago

For years, SEO meant one thing: Showing up on Google. The game has changed. Our free text-based course, Master SEO & AIO, teaches you how to rank in both traditional search and AI search so you can continue to scale as a @wordpress.org blogger or creator.

We are often asked how the ultra-violent games of cricket documented in the Icelandic Sagas lasted so many days (up to 11). The truth is there were many rest days, as modern blood transfusion and head transplants were not available. Impact players were often sourced nationwide.

There has been a weather station in Kew Gardens since 1871 a span of 155 years and counting. July 2026 was driest July in series, formerly July 1999 2.5mm and of all months with 0mm. A constant has been high variability, but we are seeing increased very dry July's.... (1/2)

Bild