Fabrice Bellard's Ts_SMS: Short Message Compression Using LLM (bellard.org)

My own tests:

     ./ts_sms c "TIL about ts_sms, an LLM based compression system for short text messages (like tweets)."
    夋鷦네樇䶙邁㟪䤣䎧随捋㜁駊

     ./ts_sms d 夋鷦네樇䶙邁㟪䤣䎧随捋㜁駊
    TIL about ts_sms, an LLM based compression system for short text messages (like tweets).

And:

    >>> len("TIL about ts_sms, an LLM based compression system for short text messages (like tweets).".encode('ascii'))
    88
    >>> len("夋鷦네樇䶙邁㟪䤣䎧随捋㜁駊".encode('utf8'))
    39

According to ChatGPT:

    The text appears to be a sequence of Chinese characters mixed with Hangul (Korean script). However, as written, it seems nonsensical or encoded and doesn’t form a coherent phrase in either language.

Impressive compression rate.

Related project: https://bellard.org/nncp/ and https://bellard.org/nncp/nncp_v2.1.pdf If any one wants to read more about how it's implemented.

Reposted later with more traction https://news.ycombinator.com/item?id=42549083

Related submission from today:

https://bellard.org/ts_zip/

https://news.ycombinator.com/item?id=42549083 (37 comments)