This one? bsky.app/profile/blue...
11.03.2026 15:23
👍 1
🔁 0
💬 0
📌 0
This one? bsky.app/profile/blue...
I left OpenAI Pro today, as I despise them jumping to a deal with the US Department of War at the point when Anthropic made a stand against certain War Department’s AI usage.
www.theguardian.com/us-news/2026...
Å då pratar vi om bio baserad grafen
Also that the Coffee saying “Made in France” is disturbing. Egoiste seems to be able to buy the capsules from UCC Coffee Europe (but that in turn seems to be owned by a Japanese mother company).
Shame on them!
Ja, och tänk vad praktiskt med en hangar där man kan sortera alla sina (för många) prylar!
Only two people have ever been gifted a Nobel Peace Prize: one was an authoritarian from a brutal fascist regime, and the other was Joseph Goebbels.
Illustrated by @paullr.bsky.social:
Well, it depends on the alternative cost of the drone hitting its target. If that is in the millions, it might make sense in the short term. But obviously, to make this sustainable, the cost needs to come down.
Ja, och på engelska snubblar jag (för) ofta på "resaerch"
Ett av sovrummen kan man bara komma till om man går ut på balkongen, känns inte så mysigt på vintern.
Maybe this was one of their relaying drones?
Very Cole looking!
/** * Integer square root using the classic "binary digit‑by‑digit" algorithm. * Implements the method described at: * https://www.cantorsparadise.com/the-square-root-algorithm-f97ab5c29d6d * The algorithm works by examining result bits two at a time, starting from the * most‑significant pair that could affect the result. */ // Integer square root using the digit‑by‑digit (hand) algorithm for unsigned long unsigned long sqrt_binary_hand(unsigned long x) { unsigned long res = 0; unsigned long bit = 1ul << 62; // start with the highest possible bit pair for 64‑bit // Align bit to highest possible position <= x while (bit > x) { bit >>= 2; } while (bit != 0) { if (x >= res + bit) { x -= res + bit; res = (res >> 1) + bit; } else { res >>= 1; } bit >>= 2; } return res; }
This is a "hand" sqrt of integers in C that codex provided, maybe useful? (Se alt text for the code)
(\_/)
( •_•)
/ 🫶 )
And for Swedish I use:
alias is='pbpaste | llm -s "Förbättra svenskan. Skriv endast ut den förbättrade texten, utan förklaringar eller inledande hälsningar" -m gpt-oss:20b | tee >(pbcopy)'
alias ie='pbpaste | llm -s "Improve English. Output only the improved text, without explanations or greetings" -m gpt-oss:20b | tee >(pbcopy)'
On a Mac, using the built in pbpaste and pbcopy, one can use @simonwillison.net’s llm to improve the text in the clipboard (in place). That is, I can now give the command “ie” in the terminal and get an improved text back into the clipboard.
Generated by Qwen/Qwen-Image, using the Prompt: A pelican riding a bicycle along a coastal path overlooking a harbor
Generated by Qwen/Qwen-Image, using the Prompt: A pelican riding a bicycle along a coastal path overlooking a harbor
Finally the "A pelican riding a bicycle along a coastal path overlooking a harbor", of @simonwillison.net fame, looks good, even if one of the feet is not on the pedal.
Generated by Qwen/Qwen-Image, using the Prompt: A 3D render of a coffee mug placed on a window sill during a stormy day. The storm outside the window is reflected in the coffee, with miniature lightning bolts and turbulent waves seen inside the mug. The room is dimly lit, adding to the dramatic atmosphere. Ultra HD, 4K, cinematic composition.
Dall-E Stormy Coffee
For another comparison to Dall-E, “A stormy coffee mug”, I think the dynamics of Dall-E output is more appealing, and I like the glass mug better than the ceramic one with a cutout.
Generated by Qwen/Qwen-Image, using the Prompt: A vintage travel poster for Venus in portrait orientation. The scene portrays the thick, yellowish clouds of Venus with a silhouette of a vintage rocket ship approaching. Mysterious shapes hint at mountains and valleys below the clouds. The bottom text reads, 'Explore Venus: Beauty Behind the Mist'. The color scheme consists of golds, yellows, and soft oranges, evoking a sense of wonder.
Dall-E Explore Venus
Comparing it to Dall-E, “Venus travel”, it seems to miss information on styles like “vintage travel poster”.
Generated by Qwen-Image with prompt: "A new year fireworks over a river with a city on the other side. Write on top 'Gott Nytt År' with 'önskar Tomas' as a subheading."
Generated by Qwen/Qwen-Image, using the Prompt: Sketches of a conceptual computer mouse as if drawn by Leonardo da Vinci, featuring various mechanical parts, exploded views, and design notes in an old manuscript style. Any text should be in latin and mirrored in the way Leonardo was writing with his left hand. Text on top of page, using cursive writing style: 'Instrumentum electromechanicale ad machinam analyticam interagendam'
For other languages like Swedish and Latin it still fails miserably, which is interesting (that the image generation of text is connected to words more than characters).
I have started to explore #Qwen-Image. Some early observations: on a Mac M4 with 128GB memory a 512x512 image takes ca 2 min, while a 1664x928 takes 30 min to generate. Its outstanding feature is the generation of English and Chinese text. The output is often really nice, but…
It seems to be able to do sqrt, logs, and trigs as well, but I have not found any documentation of this feature, so I do not know its limits.
For Mac users: now TextEdit can actually do math directly when you edit. For example if you on a line write “(4*3)+3*(4-1)-3” and then add “=” the editor will fill in “18”
Very handy!
Lite språknörderi inför kvällen med #Eurovision2025 och KAJ från Umeå Universitet:
Fyra fakta om finlandssvenska: www.youtube.com/watch?v=JecI...
Vill du låta mer som KAJ? www.youtube.com/watch?v=SsiQ...
Some code to get started: colab.research.google.com/drive/1G4_uM...
Did anyone find what they were looking for?
But maybe in some other framework and have noticed a performance difference (or not) there?
I have reported this to Keras github.com/keras-team/k... with example code, and Keras people is starting to point to Apple.
Still to go after apple, more examples would be nice to have (like M1,M2 or M3; or not using Keras). Maybe @simonwillison.net have seen something suspicious?
Has anyone with a M4 Mac seen problems or at least performance differences when running Keras3 + (TF or JAX) compared to the same code running on NVIDIA hardware? #ML #MLSky
I really like the publication of "negative" results, one can learn a lot from such "failures"! A (partially) failed attempt at improving the Transformer architecture. www.eloidereynal.com/p/a-partiall...