Run private offline ‘chat’ AIs on Windows, without wasting a day on a frustratingly complex and fiddly Python-based setup and install. The free LM Studio makes discovery, downloading and running chat LLM AI’s as easy as installing any other Windows software. A good graphics card is required to run your selected AI chat. Note also that the LM Studio wrapper requires a PC with a processor that supports AVX2, if not one with a dedicated NPU. There’s also a beta release that supports older plain AVX processors.

New on Archive.org, a long run of the journal Cuadernos Hispanoamericanos.

New on Archive.org, a recent run of the Library of Congress magazine. Appears to be 2013-2022.

Google is killing off its “cached” feature, reports Ars Technica. The feature kept a copy of a page for a few hours, days or weeks. Sometimes longer. The burden will now largely fall on permanent preservation in The Wayback Machine of the Internet Archive, making that service more vital than ever. However, that does have limitations, said to be ‘100 saves per day, per IP address’. Thus if your ISP puts you on a shared IP, you could be out of luck that day.

There’s also Archive.is, but there can be a queue 1,000 users long to archive a page. But it’s otherwise fast and also saves a screenshot. There are a few others, such as Perma.cc.

It might have been nice if Google had also bunged the Internet Archive $100m or so, to help them take up the slack, but Google seems to be a bit hard up these days. Ars Technica suggests the cache killing is a cost-saving move.

The Paul Allen Institute for AI has open-sourced its OLMo AI models for text generation. Funded by the wealth of Microsoft’s Paul Allen, the Institute runs the huge free Semantic Scholar ‘academic papers’ search harvester and database, and also has an AI arm. Its AI models are radically ‘open’ under an Apache licence, available for free-use including commercialisation. My guess would be that OLMo may be especially useful for academic text and semantics?