Technology

Full image and other similar screenshots

Ghostarchive mirror: https://ghostarchive.org/archive/R9M9R

UdenUSA is currently the fourth most downloaded app in Denmark on the App Store, the American ChatGPT is in fifth place

Google has reinstated support for the JPEG XL image format in the open source Chromium code base, reversing a decision it made in 2022 to remove it.

The update allows Chromium to recognize, decode, and render JPEG XL images directly, without extensions or external components.

This change applies at the browser engine level, meaning it will affect future versions of Google Chrome and other Chromium-based browsers when they are released.

Every year, at the end of January and the beginning of February, thousands of people from Europe and around the world gather in Brussels to discuss open source and open technologies. The main attraction is FOSDEM, Europe’s largest open source conference, which has inspired a range of side events, social activities, and workshops. For those interested in open technology, digital policy, and EU developments, OpenForum Europe’s EU Open Source Policy Summit brings together open source leaders and policymakers. Together, these events make up the EU Open Source Week.

https://archive.ph/jSRoU

Suppliers of parts for Nvidia’s H200 have paused production after Chinese customs officials blocked shipments of the newly approved artificial intelligence processors from entering China, according to a report.

Nvidia had expected more than one million orders from Chinese clients, the report said, adding that its suppliers had been operating around the clock to prepare for shipping as early as March.

Chinese customs authorities this week told customs agents that Nvidia’s H200 chips were not permitted to enter the country, Reuters reported.

Sources have also said government officials summoned domestic tech firms to warn them against buying the chips unless it was necessary.

The paper argues that we have been wasting a lot of expensive GPU cycles by forcing transformers to relearn static things like names or common phrases through deep computation. Standard models do not have a way to just look something up so they end up simulating memory by passing tokens through layer after layer of feed forward networks. DeepSeek introduced a module called Engram which adds a dedicated lookup step for local N-gram patterns. It acts like a new way to scale a model that is separate from the usual compute heavy Mixture of Experts approach.

The architecture uses multi head hashing to grab static embeddings for specific token sequences which are then filtered through a context aware gate to make sure they actually fit the current situation. They found a U shaped scaling law where the best performance happens when you split your parameter budget between neural computation and this static memory. By letting the memory handle the simple local associations the model can effectively act like it is deeper because the early layers are not bogged down with basic reconstruction.

One of the best bits is how they handle hardware constraints by offloading the massive lookup tables to host RAM. Since these lookups are deterministic based on the input tokens the system can prefetch the data from the CPU memory before the GPU even needs it. This means you can scale to tens of billions of extra parameters with almost zero impact on speed since the retrieval happens while the previous layers are still calculating.

The benchmarks show that this pays off across the board especially in long context tasks where the model needs its attention focused on global details rather than local phrases. It turns out that even in math and coding the model gets a boost because it is no longer wasting its internal reasoning depth on things that should just be in a lookup table. Moving forward this kind of conditional memory could be a standard part of sparse models because it bypasses the physical memory limits of current hardware.