This is the basic flaw in almost every generative AI use case. If quality matters at all, the quality control makes it not worth the trip.

The reverse correlation of how mgmt thinks of people vs tech amazes me.

The tech can make a ton of mistakes, but that's OK, it will magically get better someday. To hire a person they must prove they have solved your exact problem before and 100% guaranteed to be successful on day one.

A libel case in the US against a public figure requires actual malice, not malicious intent. Actual malice can include a reckless disregard for the truth. Even if it were the US, he'd still have a case

It's not really an AI issue, it's a training database issue (a classic). Which means the bias is actually human--the people who made the training database for the AI failed to provide enough samples across race (2/?)

There are always empty words about "human in the loop", that you should always check the work of an LLM. But the entire value of a summary is as a replacement for reading. Nobody uses an LLM to generate a "summary" *and* independently verifies.

I don't mean to come across as snide, but I'm tired of hearing what AI *could* do. I want to hear what it's doing that makes it worth keeping around.

Steve Jobs didn't need to go around telling people what Apple would be in 5 or 10 years. He just said what awesome thing it was doing now.

also designing and writing code is the fun part. make an AI that goes chasing down half a dozen stakeholders and gets them to agree on requirements or something.

As I said, LLM proponents tend to be dismissive whenever it's suggested that they try to measure their own productivity with vs. without LLM assistance

Thanks for the illustration, I guess?

But how will I make a link that's styled like a button without sticking the button element in increasingly bizarre places? 😭😭😭

At this point I'm convinced that modern languages should not have exceptions. Exceptions are just fancy goto. A language should depend on returning an error tagged union and have some syntactic sugar for early return of an error type. Rust and Haskell get this right.

Oh I didn't realize it was an existing meme

What?? I didn't mention anything about Tiktok. I've noticed that whenever people do polls of their own audience of what format they'd prefer to receive a tutorial or piece of information the outcome is 1. video 2. text 3. podcast. People don't like to read

That's what I meant. Because of randomization, we can more confidently attribute the difference in outcome to a difference in effect. However, that limitation is worth noting imo

<span className="text-center”>You can eliminate that problem by adopting a new, different set of problems</span>