Package: `quackingllama`. Basically, a llama that quacks like a duck (or, more technically, a package that relies on `Ollama` for interacting with LLMs, and on DuckDB for caching results). - giocomai.github.io/quackingllama/ #rstats

New post, in which I indulge in exploring categorisation of text with open LLMs, and along the way incidentally develop an R package to facilitate relevant workflows.

Analysing Russian news via Telegram, processing them with open LLMs - tadadit.xyz/posts/2025-0...

Belated announcement that I have finally released an openly-licensed collection of textual datasets extracted from Russia's main government websites.

Yearly update expected.

tadadit.xyz/posts/2024-1...

Do check out the full posts for more details on the technicalities as well as more context on how Prigozhin communicates through audio messages. tadadit.xyz/posts/2023-0...

This is the result of offline transcription, using #whisper. Quality of transcription is surprisingly good, translation somewhat less so (using e.g. Google Translate on the Russian transcription may work better).

Few posts still missing from English version, will finalise soon.

You can read a transcription of all audio messages posted by Prigozhin in the months before his mutiny in a single page at this link: tadadit.xyz/posts/2023-0...

Transcribed and translated in English here: tadadit.xyz/posts/2023-0...

Transcription of these audio files shows just how unhinged his criticism of the military leadership had become, reaching an audience of millions.

If you are not rebuffed after weeks of repeating this, you may be (figuratively) excused for thinking the Kremlin is really on your side.

There is also an obvious cutoff date, when he started to openly criticise by name Shoigu and Gerasimov (after 9 May, Victory Day)

Interesting to notice that just by looking at the number of audio messages posted each day, changepoint algorithms correctly detect key points in Prigozhin’s rhetorical escalation before his mutiny.

As Prigozhin’s messages are back in the news, here is a lengthy methodological post dedicated to the 400+ audio files that #Prigozhin posted before his mutiny in June 2023.

Full transcriptions provided, easy to read and download. tadadit.xyz/posts/2023-0...

Some quick results: