GoKawiil
Tech News
USB-C offers better convenience thanks to its reversible design and widespread use across laptops, tablets and smartphones, making it ideal for transferring large files on mobile devices. However, the connector type alone does not determine speed—actual performance depends on the USB specification (like 2.0 or 3.2) built into the drive and the port it's plugged into.
A newly published technical handbook explains FlashAttention, the exact attention algorithm that speeds up Transformer models by minimizing memory traffic rather than reducing computation. It walks through the standard attention implementation's repeated reads and writes to GPU high-bandwidth memory, then details how tiling, online softmax, and recomputation eliminate most of that overhead across FlashAttention-1 through FlashAttention-4.
External SSDs use flash storage that speeds up file transfers and editing directly from the drive, while offering greater durability, quieter operation, and lower power draw thanks to having no moving parts. External hard drives remain the cheaper option per gigabyte, with examples like a 2TB Seagate HDD costing less than a 1TB Samsung T9 SSD.
Comparing a 256GB SSD to a 1TB hard drive isn't a simple capacity contest, since SSDs retrieve data electronically from flash memory while HDDs rely on spinning platters and moving read/write heads, making SSDs significantly faster for booting and running programs. Whether 256GB is enough depends heavily on usage: it suits light users who rely on cloud storage, while heavy local file storage, creative work or gaming may require 512GB, 1TB or more.
Emil Barr, who launched Step Up Social from his college dorm room at Miami University, earned his first $1 million within 14 months of starting the business at age 19. Now 23, he estimates his personal net worth at roughly $35 million and has since founded a second company, Flashpass, aimed at addressing AI-driven job displacement.
Researchers behind the dealignai project published a modified version of DeepSeek's V4.1-Flash model with its safety refusal circuitry surgically removed at the weight level, while preserving core capabilities like reasoning, vision, and its 1M-token context. The team says the checkpoint loads like a standard model with no special code needed, and claims HarmBench testing shows a 100% attack success rate for harmful prompts across both low and high reasoning effort settings, compared to the base model's much lower compliance rate.
A tech writer draws on their early-2000s experience at a Manhattan creative agency, recalling the wild inventiveness of the Flash era of web design, to argue that AI's ability to easily replicate existing work has shifted the real competitive edge for businesses toward original ideas rather than features or pricing. The piece uses Flash-era microsites, games and experimental art projects as examples of a period defined by genuine creative risk-taking.
Nepal's army is using four China-donated DJI FlyCart 100 drones to make 10 to 16 daily aid runs following the August 26 glacier collapse that triggered flash floods along the Nepal-Tibet border. The drones, each capable of hauling around 132 pounds per trip, have delivered food supplies and also retrieved bodies from areas inaccessible to ground crews. Nepal's disaster authority reports 1,369 deaths and more than 5,000 people still missing as of September 9.
Elegoo is holding an Elegoo Day sale throughout September, offering its Centauri Carbon 2 Combo printer for $399, which includes 1KG of PLA filament. The Core XY printer supports a four-color Canvas system, auto bed leveling, 31 sensors, and a 256mm cubed build volume, with print temperatures reaching 350C for materials like PETG and ABS.
Luxury housing prices have declined for 29 consecutive months across every high-end price tier, according to new market data. Despite the extended price slide, luxury homes are now selling faster than they did a year ago, suggesting buyer demand remains resilient even as sellers accept lower prices.
DeepSeek has released V4.1-Flash on its API, replacing the earlier V4-Flash and V4-Flash-Vision-Exp models. The new model adds native multimodal capabilities and can be accessed by setting the model parameter to deepseek-flash.
A follow-up experiment tested four open models—DeepSeek V4 Flash, Inkling, Kimi K3, and Qwen3.8 A95B—by inserting the first 1% of GPT-5.5 Pro's reasoning trace into each model's own reasoning channel before letting it generate answers freely. Researchers then measured how much of GPT-5.5 Pro's visible answer text overlapped with each model's output. Qwen3.8 showed the largest jump, with overlap rising from 33.92% unprefilled to 54.50% with the GPT-5.5 Pro prefill, a 20.58 percentage-point increase, while other models showed much smaller shifts.