lukechilds on Nostr: GPT-4 is very capable, the resulting bash scripts work well about 95% of the time. ...
GPT-4 is very capable, the resulting bash scripts work well about 95% of the time. However there are often subtle bugs or edge cases that aren’t handled unless you explicitly tell it to look out for them.
After a few iterations of adding extra details to handle edge cases you can get something very high quality.
Results are much worse on open source LLMs but they’re catching up quickly.
Published at
2023-08-08 08:49:05Event JSON
{
"id": "f5e83b8e799a33b994c5c85f455a2c7a8fbe000c78cbfb9ee4425a1e28fdef11",
"pubkey": "bae77874946ec111f94be59aef282de092dc4baf213f8ecb8c9e15cb7ed7304e",
"created_at": 1691477345,
"kind": 1,
"tags": [
[
"e",
"171fe4a023749d0b8bc5d4e36cf0005ef1f63790f5b452d69466325227d49699"
],
[
"e",
"0044fb37547bbfae3abe1cd2d355342304f29a88202d0977dc946b569d16529f"
],
[
"p",
"330fb1431ff9d8c250706bbcdc016d5495a3f744e047a408173e92ae7ee42dac"
]
],
"content": "GPT-4 is very capable, the resulting bash scripts work well about 95% of the time. However there are often subtle bugs or edge cases that aren’t handled unless you explicitly tell it to look out for them.\n\nAfter a few iterations of adding extra details to handle edge cases you can get something very high quality.\n\nResults are much worse on open source LLMs but they’re catching up quickly.",
"sig": "177fa0048cc92a91f0444b697c73e24ecb3e8b6337e7a0a3aecb8e3958710eebe3b6b62265b9460dfb3a41bb8296754ad592cffd1bbd3b1a49bd7dd255dd35ca"
}