rcvassallo83

LLMs don't make decisions

They generate text which resembles reasoning and may include a tool call

The harness runs tool calls

The human made the decision to write the harness

The human is responsible

Am I the only one who's no longer impressed by demonstrations that one turning machine can emulate another?

Efficient window unit?

Best of the best is about 15-16 SEER

That's entry level central HVAC efficiency

Minisplits are far higher, 20+

Oof leader of bubble are starting to take a step back?

This resonates

I've had success with greenfield code followed by frustration when asking for changes to that code due to over editing

And prompting for "minimal changes" does keep the edits down. In addition to this instruction, adding specifics about how to make the change and what not to do tends to get results I'm looking for.

"add one function that does X, add one property to the data structure, otherwise leave it as is, don't add any new validation"

In my experience, the narrative the prosecutor argued on behalf of the accuser was obviously false because specific key facts were contradicted by bodycam statements, for example

"He took my phone and it was dead" -> bodycam showed her using the phone when police arrived

I provided a recording of my accuser clearly being drunk, aggressive, threatening me while I was de-escalating. I was the one who called 911 to stop her from drunk driving. Her speech clearly slurred.

Instead of realizing her story doesn't add up, the prosecutor brought in a DV expert to explain how it's typical for abusers to call 911 and that her behavior was a normal reaction to being assaulted.

Thankfully, the jury knew better.

Scary process is an understatement, especially because I was facing a domestic violence charge.

Long story short, emotional abusive partner got drunk and verbally combative, despite my attempts to de-escalate. When nothing worked I went into my bedroom and locked the door. She started pounding on the door and demanded her things. I gave them to her and told her she needs a ride home she no longer welcome. She verbally abused and provoked me for 10 minutes before getting in her car. Took the keys, called 911. She grabbed me causing us both to fall a few minutes before the cops arrived and told them I threw her to the ground. We both had a couple scapes so they arrested us both.

Interfered with the 911 call, filed a false police report, assaulted me, caused property damage. She got charged with class c assault only and a dismissal. I felt like I was seen as guilty until proven innocent.

Fortunately I recorded all her verbal abuse (prosecution tried to use it against me and brought in DV expert to explain both her conduct and my 911 call as typical in IPV cases)

Fortunately the jury didn't buy it. I was literally being threatened with violence in my own home for telling her to leave. Between that and the bodycam statements contradicting her testimony I was shocked that they didn't drop the charges or offer a favorable plea deal.

The judge was absolutely fair, prosecutor bent on punishment, alleged victim was attempting to ruin my life (as captured in my audio)

Whew!

In the end the claims were so obviously fabricated that my attorney made no defense. It was clear that the accuser was not credible.

Perjury was provable. No consequences for her. This happened in Brazoria county

The thing about the legal system is there's no incentive to investigate to find the truth.

The incentive is to prosecte and prove the charges.

Speaking from the experience of being falsely accused after calling 911 to stop a drunk woman from driving.

The narrative they "investigated" was so obviously false, bodycam evidence directly contradicted multiple key facts. Officials are interested only seeking to prove the case. Thankfully the jury came to the right verdict.

Curious, what did you do for this customer? Are they technologically sophisticated? Do they have a lot of users for what they built, or are they building tools used internally by a handful of people?

For simple tasks and code that doesn't change much over time, programmers don't add a lot of value over what an LLM can provide. It's easy to prompt your way to a good-enough tool for a small and exclusive set of users. Managing complexity beyond that is where a real programmer provides value.

Experienced developers provide a deep understanding of what's possible, knowing what to build, how to avoid pitfalls, how to adapt to new requirements. The expertise is in managing projects that evolve over time, meet the needs of hundreds to thousands of users, interact with external systems, have meaningful compliance, performance, and security requirements, etc.

Thanks for the reminder

Its early for April fools

As someone who has written a few deeply personal articles with LLM assistance, I see the signs and I'm almost certain this was generated off a few bullet points. The repetition and cadence strongly resembles the LLM output. Its the kind of fluff that I remove from a piece, because it lacks humanity and offers little substance.

Article definitely has an AI writing style

Thinking harder than I have in a long time with AI assisted coding.

As I'm providing context I get to think about what an ideal approach would look like and often dive into a research session to analyze pros and cons of various solutions.

I don't use agents much because it's important to see how a component I just designed fits into the larger codebase. That experience provides insights on what improvements I need to make and what to build next.

The time I've spent thinking about the composability, cohesiveness, and ergonomics of the code itself have really paid off. The codebase is a joy to work in, easy to maintain and extend.

The LLMs have helped me focus my cognitive bandwidth on the quality and architecture instead of the tedious and time consuming parts.