jq: How to slurp mid-filter (convert separate json objects to an array)

I am parsing Elastic logs that look like

$ cat my_log_file.txt:{"level": "INFO", "message": "foo"}{"level": "WARN", "message": "bar"}{"level": "WARN", "message": "baz"}

Because they're one per line, formerly I have used jq -s to slurp them into an actual array that I can run map on:

jq -s 'map(.message) | unique' my_log_file.txt

Now I want to select out only lines that have level != "INFO". I should be able to just use this cookbook recipe, but again jq is having trouble with each line being a separate object and not in an array.

I can't use slurp, because it doesn't work with this command:

jq 'select(."level" != "INFO")' my_log_file.txt

But when I want to map to .message again, I get the same error I got before when I wasn't using slurp:

$ jq 'select(."level" != "INFO") | map(.message) | unique' my_log_file.txtjq: error (at <stdin>:90): Cannot index string with string "message"

How can I convert my records midstream -- after the select is done, to convert the result from

{"level": "WARN", "message": "bar"}{"level": "WARN", "message": "bar"}

to

[  {"level": "WARN", "message": "bar"},  {"level": "WARN", "message": "bar"}]

I heard that inputs was designed to replace slurp, but when I tried

$ jq 'select(."level" != "INFO") | [inputs]' my_log_file.txt

that ignored my select and I had a list of every message again.