Defaults to that out of box, but you can pick a few models via a command line flag.
@patio11
-
Managing LLM Context Windows: Input Truncation Strategy
By
–
"Why did you pipe the output of the git commands to head above?" Because LLMs have limited context windows and if you exceed the length of them this tool fails gracefully, so by habit I chop the input to it down prior to asking to process it.
-
LLM CLI Tool Available on GitHub and PyPI
By
–
If you want to play with it, available on pip / Github / etc. Requires an OpenAI account for batteries-included operation or you can install an LLM locally https://
github.com/simonw/llm -
Using LLM CLI for Linux Command Generation and Automation
By
–
You can, of course, imagine many things you could do with this based on whatever you do in the terminal most frequently. A thing I find myself gravitating to immediately is acting as superman: $ llm 'Linux one liner for current date in most commonly used U.S. format please.'
-
Using Git diffs with LLM to summarize commit changes
By
–
git show 57ee | head -c 4000 | llm -s 'This is the first part of a diff. What did this commit actually *do*? Your answer should fit in a tweet.'
-
Simon Willison’s LLM command line tool demo
By
–
You might not know about @simonw
's command line 'llm' tool. Here let me demo it. git log | head -n 200 | llm -s "Of the most recent 5 commits, which is probably the most important? I use 'Minor' and similar commit messages to mark unimportant commits." -
Verification Easier Than Generation in AI Problems
By
–
This problem has the character where answers are easier to check than they are to generate, right.
-
Designing AI systems for honest users versus criminal threats
By
–
And so does that process need to be designed to optimize for the needs of its typical user, the honest person, or its typical user, a criminal you can’t offer services to? Yes.
-
Criminals vs Honest Users: Account Security Scale Problem
By
–
“Surely, criminals are not more numerous than honest working folk.” Of course they aren’t. But if you have an account opening process, your typical honest customer will go through that precisely once, and a criminal might try to go through it 13,000 times. On Tuesday.
-
Bureaucratic Process Design and Abuse Prevention Challenges
By
–
A persistent surprise for entrepreneurs and policymakers, which is illegible to customers and largely outside of their mental models, is that bureaucratic processes will disproportionately need to engage with people attempting to abuse the process.