When it comes to AI services, you don't necessarily get what you pay for. It turns out that AI models with expensive tokens may cost less than models with cheap tokens for particular tasks. And the tooling attached to those models can have a significant effect on cost and output quality. Databricks, which sells data analytics software and services, recently devised an internal coding benchmark to assess the tradeoff between price and performance using various AI models. Matei Zaharia, CTO of Databricks and associate professor of computer science at UC Berkeley, said the company undertook the evaluation because models are often tuned to existing benchmark tests like SWE-Bench – which is "broken," according to OpenAI. Databricks devised its benchmark using real engineering tasks performed by its staff to assess how AI agents perform. Zaharia said while the results reflect the company's internal codebase, other companies should be able to conduct similar evaluations using their own code. One of the things Databricks found was that open weight models like Z.ai's GLM 5.2 are competitive with frontier models, like Anthropic's Opus 4.8. "It landed in the top capability tier, statistically tied with Opus 4.8 on quality, but costing $1.28/task against Opus’s $1.94," the company said in its report. But the price-per-token doesn't tell the whole story. Databricks contends that price-per-task needs to be considered. "Cheaper per-token does not imply cheaper per-task," said Zaharia in a social media post. "For example, Sonnet 5 costs less per token than Opus 4.8 but used more tokens, resulting in higher cost and lower quality." So while Anthropic's Sonnet 5 was around 1.7x cheaper than Opus 4.8 on a per-token basis, it was more costly on a per-task basis – $2.09 for Sonnet 5 compared to $1.94 for Opus 4.8. That's because it completed tasks less often (81 percent compared to 87 percent), and consumed more tokens to achieve the desired result. Academics already reached this conclusion, noting back in March that in about a third of the model comparisons they conducted, the model with the lower listed price ended up costing more. "For example, Gemini 3 Flash's listed price is 80 percent cheaper than GPT-5.4's, yet its actual cost across all tasks is 38 percent higher," they observed. The other thing that had a significant impact on test results was the harness – software like Claude Code, OpenAI Codex, and the Pi coding agent – which passes user input to the model, invokes various tools, and returns results. "Harnesses make a huge difference in cost-performance," said Zaharia. "The very simple Pi harness got the same success rate as harnesses from the LLM vendors with Opus and GPT 5.5, but at 2x less cost!" Zaharia attributed the difference to the size of the input – the context – passed to the model with every turn. When Claude Code served as the harness for Opus 4.8, Databricks measured a context of 742,000 tokens per task, compared to 236,999 for Pi. That's about 3.2x fewer tokens overall. With Codex, the total context per task was 1,235,000 tokens, compared to 665,000 tokens for Pi, which is known for its minimal system prompt. Zaharia said the results explain why Databricks built a tool called Omnigent to harness the harnesses – it's a wrapper for combining and swapping multiple coding agents. It's the front-end equivalent of the kind of back-end model swapping that OpenRouter enables. ®

Thomas Hawk posted a photo:

Thomas Hawk posted a photo:

Al sinds zijn verkiezingswinst in april wilde premier Péter Magyar dat president Tamás Sulyok aftrad. Sulyok, die door Magyar werd bestempeld als een „marionet” van de radicaal-rechtse oud-premier Viktor Orbán, weigerde dat.

Microsoft geeft als eerste grote Amerikaanse partij inzicht in het stroomverbruik van zijn datacenters in Nederland. Dat is groot, en groeit de komende jaren bovendien hard. Terwijl andere datacenteruitbaters zwijgen, verhardt de strijd om toekomstige aansluitingen op het stroomnet.

Nederland waterland. Dat ligt niet alleen ontzettend lekker in het gehoor, het ís ook nog eens zo. MAAR VOOR HOELANG NOG? Al tijden wordt gewaarschuwd voor grote watertekorten en zelfs De T. zucht, maar nu brengt Nieuwsuur ons een waarschuwing rechtstreeks van de Landelijke Coördinatiecommissie Waterverdeling, die stelt dat we waterwise zo'n beetje teruggaan naar het jaar 1976 (toen Joop den Uyl nog kraanbeheerder was). We zitten pas een paar dagen in fase 1: dreigend watertekort, maar fase 2: feitelijk watertekort ligt al op de loer. Dat raakt in de eerste plaats scheepvaart en boeren, alleen op een gegeven moment komt ook de burger in de zee van ellende terecht, zeker met die eeuwigdurende hittegolf waar we in zitten. En ja, mogen we dan straks nog wel dweilen? Plantjes water geven? De was doen in een teil? Kan er dan nog muntjesloos gedoucht of moeten we de zweetgeur van Jeroen van sales op kantoor gaan gedogen? Dat worden onsmakelijke toestanden in Nederland. Gelukkig kunnen we in plaats van drinkwater nog altijd aan het bier (toch geldt ook hier: hoelang nog?), verder ziet het er allemaal bepaald niet positief uit.

Goed genuanceerd verhaal wel

Social

Jeongmin Lee is interested in the ways “memory is carried through craft and repetition.” On traditional Korean mulberry paper, or hanji, Lee draws delicate lines in ink and pigments known as bunchae, rendering rippling textures that whirl across the page. Steeped in local folklore and mythology, the Busan-based artist creates surreal scenes that conjure fantastical tales of life by the sea.

“Most of my recent projects begin with reading regional folktales, visiting places connected to those stories, and collecting fragments of history, mythology, and oral traditions,” she says. “I rarely paint a folktale exactly as it’s written; I’m more interested in its symbols, emotions, and the questions it leaves behind.”

Many of Lee’s compositions focus on women’s knowledge, labor, and resilience and how those qualities emerge through coastal storytelling. Tales of powerful sea gods and the diving traditions of the haenyeo commingle into balanced illustrations that translate the symbols and motifs of the region anew.

While her approach is labor-intensive, Lee enjoys utilizing such a meticulous, meditative technique. “Painting with traditional pigments requires a slow, layered process. The colors are built gradually, allowing the paper and pigments to create subtle textures that wouldn’t be possible with other materials,” she adds. “It gives me space to sit with these stories while I paint them.

Many of the works shown here are part of Daughters of the Sea, an ongoing series recently on view at the SĀBRS Festival in Riga. Lee hopes to expand the project into one that’s participatory and connects similar narratives from across the globe. “I’m exploring ways that folklore can become something people experience and talk about together, rather than something that exists only on a gallery wall,” Lee shares.

She’s also currently working on a graphic novel centering on Busan’s mythology and folklore, which she hopes to complete next year. Keep up with her projects on Instagram.

Do stories and artists like this matter to you? Become a Colossal Member today and support independent arts publishing for as little as $7 per month. The article Korea’s Coastal Folklore Surfaces in Jeongmin Lee’s Ink Illustrations appeared first on Colossal.

I have a Pi 4 B that, after a power failure, seems unable to talk HDMI in any resolution higher than 1024x768. I have tried both ports, multiple cables, multiple monitors, and a known-good CF card.

# xrandr Screen 0: minimum 320 x 200, current 1024 x 768, maximum 7680 x 7680 HDMI-1 connected primary 1024x768+0+0 (normal left inverted right x axis y axis) 0mm x 0mm 1024x768 60.00* 800x600 60.32 56.25 848x480 60.00 640x480 59.94 HDMI-2 disconnected (normal left inverted right x axis y axis)

I have also tried various permutations of hdmi_force_hotplug, hdmi_group, hdmi_mode, hdmi_force_mode, hdmi_force_hotplug and hdmi_drive in config.txt, and video=HDMI-A-1:1920x1080@60D in cmdline.txt with no change.

Before I toss it in the trash, does anyone know what could have caused this and if it is fixable? Is this a known failure mode of the video hardware scorching itself somehow?

Do not speculate. Do not just google it for me.

Previously, previously.

Ik leerde Tiong Ang kennen via Remy Jungerman met wie hij in 2002 een residentie deed in Indonesië. Ik ben zijn werk sindsdien blijven volgen, eerst toen ik meer reisde op afstand en sinds ik [Meer...]

An anonymous reader quotes a report from The New York Times: A state-owned newspaper in China recently published a satellite image of a data center in Gainesville, Va., writing in English that the development of artificial intelligence posed a threat to Americans' physical and financial well-being. A comic strip made to look as if it had been published by a Maryland news outlet -- created with OpenAI's ChatGPT by people in China, the tech company said -- circulated on X this year, blaming data centers for soaring electricity bills. It showed a tycoon smoking a cigar and clutching bags of cash. A video shared on X by a known covert Russian influence operation questioned the viability of a data center that an American company, Firebird, is constructing in Armenia, the small Caucasus nation that has been a focus of Kremlin pressure. "The country's electrical grid instability may render it useless," the video's narrator says.

All are examples of a push by foreign adversaries to seize on what polls have shown is deep ambivalence -- verging at times on hostility -- about the spread of the data centers needed to power A.I. in the United States and elsewhere. China, Russia and, to a lesser extent, Iran have sought to use state media outlets to turn the controversy over data centers in the United States into "a domestic fracture point," according to a new analysis by Alethea, a threat intelligence company, which identified scores of articles and posts on social media this year. These campaigns, whose impact on public opinion remains to be seen, have raised alarms in Washington, where A.I. is seen as a top issue heading into this year's midterm elections.

Read more of this story at Slashdot.