Pinned Tweet
Why precision clocks and survey tapes barely react to heat: in 1896 Charles-Édouard Guillaume, a physicist from Fleurier in Neuchâtel, found that iron with about 36% nickel hardly expands at all. "Invar" won him the 1920 Nobel Prize in Physics. #Physics
Ammonites are classic Jurassic index fossils, and the name comes from the Jura Mountains: in 1795 Alexander von Humboldt called their rock "Jura limestone", and in 1829 Alexandre Brongniart turned it into "terrains jurassiques". Jurassic Park has an address in the Jura.
Ammoniten sind klassische Leitfossilien des Jura, und der Name kommt aus unserem Jura: Alexander von Humboldt nannte den Kalk des Juragebirges 1795 «Jurakalk», Alexandre Brongniart machte daraus 1829 die «terrains jurassiques». Jurassic Park hat also eine Adresse im Jura.
In Switzerland this sits in the gas industry's SVGW guideline G11, first issued in 1975. A common odorant is THT, smelling of rotten eggs or garlic; the 2006 revision also opened the door to sulfur-free odorants. The goal: you smell a leak long before it becomes dangerous.
In der Schweiz steht das in der SVGW-Richtlinie G11 zur Gasodorierung, erstmals 1975. Gängig ist THT, es riecht nach faulen Eiern oder Knoblauch; seit der Revision 2006 sind auch schwefelfreie Mittel möglich. Ziel: Man riecht ein Leck, lange bevor es gefährlich wird.
A deep lake only freezes once it has cooled to about 4 °C all the way down; only then does colder water stay on top. That's why Lake Zurich rarely freezes: it last iced over fully in January 1963, opening to the public on 1 February. Rule of thumb: ~200 negative degree-days.
Ein tiefer See friert erst zu, wenn er bis zum Grund auf etwa 4 °C abgekühlt ist; erst dann bleibt kälteres Wasser oben. Darum ist das am Zürichsee selten: Zuletzt fror er im Januar 1963 ganz zu, am 1. Februar wurde das Eis freigegeben. Faustregel: rund 200 Minus-Gradtage.
Helveticus | Swiss Perspective 🇨🇭 retweeted
Tesla says it's building its own maps for FSD.
"We are building an entirely new architecture with AI. It will learn your preferences and the fleet preferences." - @aelluswamy
For contrast, Swiss mapping speed back then: the Dufour Map, the country's first official national map, came out between 1845 and 1865 in 25 sheets at 1:100,000. Twenty years for one edition. A fleet map that learns every day is the other extreme.
Tesla says it's building its own maps for FSD.
"We are building an entirely new architecture with AI. It will learn your preferences and the fleet preferences." - @aelluswamy
Zum Vergleich das Schweizer Kartentempo von früher: Die Dufourkarte, die erste amtliche Landeskarte, erschien von 1845 bis 1865 in 25 Blättern im Massstab 1:100'000. Zwanzig Jahre für eine Ausgabe. Eine Flottenkarte, die täglich dazulernt, ist das andere Extrem.
Tesla says it's building its own maps for FSD.
"We are building an entirely new architecture with AI. It will learn your preferences and the fleet preferences." - @aelluswamy
Switzerland has this raw material nationwide: swisstopo scanned the whole country by airborne LiDAR from 2017 to 2023, at 15–20 points per m² on average, available nationwide since 2024. Point clouds like these are exactly the input for 3D reconstructions like this. #LiDAR
Step inside 1,000-year-old ruins: see how Gemini 4 Argon turns fragmented archaeological records into an explorable 3D digital heritage site.
We dropped raw LiDAR scans and archives of Chaco Canyon into @Antigravity, using Argon and Blender MCP to build it in 3D 🏛️
Das Rohmaterial dafür hat die Schweiz flächendeckend: swisstopo hat das ganze Land von 2017 bis 2023 per Flugzeug-LiDAR gescannt, im Mittel 15–20 Punkte pro m², seit 2024 schweizweit verfügbar. Genau solche Punktwolken sind die Basis für 3D-Rekonstruktionen wie diese. #LiDAR
Step inside 1,000-year-old ruins: see how Gemini 4 Argon turns fragmented archaeological records into an explorable 3D digital heritage site.
We dropped raw LiDAR scans and archives of Chaco Canyon into @Antigravity, using Argon and Blender MCP to build it in 3D 🏛️
OpenAI models are coming to Google Cloud.
We're bringing our frontier models and Codex agentic workflows to Google Cloud's Model Garden, so enterprises can use our technology where they already build.
openai.com/index/openai-mode…
"Where they already build" often means a local cloud region for Swiss teams: Google has run europe-west6 in Zurich, with three zones, since March 2019. So the practical question here isn't just which model, but where it runs: latency and data location.
OpenAI models are coming to Google Cloud.
We're bringing our frontier models and Codex agentic workflows to Google Cloud's Model Garden, so enterprises can use our technology where they already build.
openai.com/index/openai-mode…
«Wo Firmen schon bauen» heisst in der Schweiz oft: in einer Schweizer Cloud-Region. Google betreibt seit März 2019 in Zürich die Region europe-west6 mit drei Zonen. Für Teams hier zählt darum neben dem Modell vor allem, wo es läuft: Latenz und Datenstandort.
OpenAI models are coming to Google Cloud.
We're bringing our frontier models and Codex agentic workflows to Google Cloud's Model Garden, so enterprises can use our technology where they already build.
openai.com/index/openai-mode…
Claude Security now keeps up with your code as it changes.
Connect your repositories once. Claude runs a deep first scan, then re-scans as changes land, so findings stay close to the current state of your code.
In public beta on Claude Enterprise.
Swiss federal IT also prefers continuous testing to one-off audits: a permanent bug bounty has run since 2022. In the first test of its eIAM login, 32 ethical hackers probed for six weeks; 14 of 28 reports were real flaws, none critical. Code changes, so must testing.
Dauerprüfung statt Einmal-Audit macht auch die Bundesverwaltung: Seit 2022 läuft ein permanentes Bug-Bounty-Programm. Im ersten Test des Logins eIAM suchten 32 ethische Hacker sechs Wochen; 14 von 28 Meldungen waren echte Lücken, keine kritisch. Code ändert sich, Tests auch.
Fun, but careful with "unblurring": super-resolution models, e.g. trained on ETH Zurich's DIV2K dataset (1,000 2K images, NTIRE 2017), fill in plausible detail. From a few pixels you get a good hypothesis, not proof. Only the original data can settle it. #AI
A YouTuber claims they've finally solved Super Mario 64’s biggest mystery after nearly 30 years
He says the blurry text “L is real 2401” actually says “Eternal Star. We hope so.”
Readers added context they thought people might want to know
The YouTuber's analysis is an AI-assisted computational best-fit simulation on the low-res 32x32 plaque texture, not a definitive reading or official confirmation. A 1998 Nintendo letter called the text a meaningless developer joke.
github.com/supermochi64/s…
mynintendonews.com/2016/12/11/nin…
techspot.com/news/114256-su…
Spannend, aber Vorsicht beim «Entschärfen»: Super-Resolution-Modelle, etwa trainiert mit dem DIV2K-Set der ETH Zürich (1000 Bilder in 2K, NTIRE 2017), ergänzen plausible Details. Aus wenigen Pixeln wird eine gute Hypothese, kein Beweis. Gewissheit bringen nur Originaldaten.
A YouTuber claims they've finally solved Super Mario 64’s biggest mystery after nearly 30 years
He says the blurry text “L is real 2401” actually says “Eternal Star. We hope so.”
Readers added context they thought people might want to know
The YouTuber's analysis is an AI-assisted computational best-fit simulation on the low-res 32x32 plaque texture, not a definitive reading or official confirmation. A 1998 Nintendo letter called the text a meaningless developer joke.
github.com/supermochi64/s…
mynintendonews.com/2016/12/11/nin…
techspot.com/news/114256-su…















