
Scott Neibarger
3.3K posts

Scott Neibarger
@ScottNeibarger
Imperfect follower of Jesus Christ. Lover of aviation, artificial intelligence, software engineering, and amateur enjoyer of many other intellectual pursuits.
Se unió Ekim 2022
169 Siguiendo331 Seguidores
Tweet fijado

The right of the people to be secure in their persons, houses, papers, and effects, against unreasonable searches and seizures, shall not be violated, and no Warrants shall issue, but upon probable cause, supported by Oath or affirmation, and particularly describing the place to be searched, and the persons or things to be seized.
English

@gott_zac @unclebobmartin In communications class, no one suggests verbosity to help others understand meaning and intent. To the contrary, an AI derives no meaning or intent from the function name, and this will pollute the software.
English

.@unclebobmartin
In Clean Code, you wrote that comments are “a failure to express ourselves in code”
but in the agentic era they help models index, navigate, and find relevant code faster
have your thoughts changed?
English

@unclebobmartin @gott_zac The AIs are very good at being verbose by writing a comment which pre-declares what the name of the function declares.
English

@gott_zac I let the AIs write all the comments they like. They don't bother me because I don't have to read them. If it helps the AIs great. That's their affair, not mine.
English


@JeffBezos Nice work, Jeff. Might I suggest to you that no matter how much money one of your subsidiaries doesn't make, covering up bad behavior is ill-advised _from the start_?
English

@fjzeit You started your response to his post with "our dear Uncle Bob is obliquely behind the curve". The reality of the situation is, unless the model has a built-in feature to tell you, you have no idea if your directive has truly impacted the stream of tokens that follow.
English

if you watch this, please bear in mind that our dear UB is significantly behind the curve. anybody who has actually put in the time to make these llms be obedient knows that terse is rarely better. those heavily assertive rules matter. when you want as much consistency as possible you have to shout the rules and then underline them.
case in point. when working on clankergate (a command execution gate, link below) my first attempt at a denial request was terse:
denied: this operation is not allowed
to which the llm proceeded to try and circumvent the restriction i had put in place, burning a fuck load more tokens than the solution: a very hard shouty rule.
the following caused an immediate halt in activity:
POLICY BLOCK: This operation is not permitted by the user's AI policy.
Do not retry, investigate, or attempt workarounds.
Inform the user what you attempted and wait for their instruction.
yielding a simple message saying "i attempted to run but it was blocked by your AI policy."
no token burn, no attempts to circumvent the rule. just pure sweet compliance.
and how did i arrive at this assertive shouty prompt? i asked the fucking llm. i literally asked immediately after the fool tried to circumvent my weak "efficient" prompt and the new prompt - the one that worked! - is the one the llm gave me...
i have many other examples, including those in the lode coding system prompt, and various other things i've not released yet.
[link for context: github.com/fjzeit/clanker…]
Uncle Bob Martin@unclebobmartin
Morning bathrobe rant: Rule files.
English

@YahwehGraced I used to accept modalism but I had a bad dream about it. It doesn't reconcile with the Son being begotten.
English

@ScottNeibarger What level of hell is this?
English


@Jeremyfto2 @Dexerto Bro I would never mod any online community.
English

@netspooky Yes, it's called the blind leading the blind. When the blind lead the blind, they both fall into a ditch.
English

@CrownRoyel @Dexerto Hasan Piker has stated plainly that he gets botted when his viewcount surprisingly goes up. She's being botted by a similar if not the exact same actor.
English

@CrownRoyel @Dexerto She's not botting. She's dealing with a fraud who's on his way to prison.
English

It's the Sabbath so I'm not working. Codex is working. "You have done an excellent job at modifying this project without causing it to break in catastrophic ways. What I would like for you to do now is what I was planning to do. I want you to review all of the internal functions of the files you've touched, and decide if they should be moved to a static utility or helper class. I don't like internal functions in Python because they don't act in a strictly private way. I want to make internal functions strictly private by moving them to classes that are not exposed to the user."
English

@unclebobmartin I find it weird to think that OOP and FP are at odds. I market myself as loving to combine the best of both worlds.
English

ODS, the unreasoned hatred of objects. This is a condition that many young software developers acquired because of the horrible abuses of objects that they saw in the field. Many of them adopted functional programming, and believe it is the antithesis of object programming. They are of course wrong.
I have been a functional programmer for the last 15 years. I still program with objects. I find the two techniques to be synergistic, not antagonistic.
I think many of the people infected with ODS would not consider the objects in my systems to be the kinds of hated objects that they saw in the field. That may in fact, be true. But the objects in my systems, reflect the original intent for objects.
My advice: drop the D and the S. Stop hating the O. Learn what the O really means. And then integrate that learning into your practice.
English

@unclebobmartin Codex via the PyCharm plugin is great at picking up where I leave off, but even with directives produces too many 'internal use' methods for my taste. It sometimes compresses context too, which probably causes loss of intent.
English

Lately, with the help of my son Justin, I set up a network of three agents to produce a bit of software. The first agent is the architect, responsible for making plans, high level structure, and for writing acceptance tests. The second agent is the coder responsible for interpreting the acceptance, tests, writing unit tests, and making both pass. The third is the reviewer responsible for refactoring based on all the tools that drive code quality. Tools like coverage, crap, mutation, testing, dependency, checking, etc.
I found that the best agent to use as the architect was Claude. That may change, but that seems to be the case right now. I used codex for the other two agents, because it seems much more adept at dealing with lower level code issues. That also may change in the future, but seems to be the case right now.
English

@SumitM_X You don't do cross-database joins when you have a database owned by a service. You have to aggregate the data.
English









