Comment by Nullabillity
3 days ago
> There's a weird attitude on this site towards AI: if it's for coding or science, people generally recognize AI tooling as effective, although imperfect, and rapidly improving.
I suspect that this is more of a selection bias thing. AI is garbage everywhere, but "AI in tech" posts tend to be hopeless abysses that are not even worth engaging with at this point. Hence, only the hucksters and grifters remain in there.
This is the camp I'm in. I've given AI the "college try", I've tried using in my workflows, and I've found that there are some cases where it genuinely has helped. But there is far too much drivel and hype.
I want to hear more from the people who've embraced it for a year, found it's pitfalls and perks, and reflect on it. I'm tired of the treadmill of content from someone who signed up for OpenAI on a Monday, used it for a JIRA ticket on Tuesday, then rushed to belt out a blogpost about how their career is forever changed on Wednesday.
Maybe we need an AI to filter out the bland bullshit content created by AI.
I swear every single post LinkedIn highlights to me is the same AI template.
Why do you need an AI to help you delete your Linkedin account?
> But there is far too much drivel and hype.
Absolutely, unarguably true, for this and every other tech boom.
But it's not all drivel and hype. There's some genuinely useful tooling here. For businesses, document summarization, translation, and asking questions about a corpus of documents using natural language are a few. For coding, some level of improved auto complete up to complete code generation are use cases. For science, there's a ton of automated testing, pattern recognition, vision based recognition use cases. For 3d graphics, where I work, some version of Nerfs could revolutionize parts of the field (although it's too early to tell) while AI based upscaling, frame generation, and path tracing noise removal are already causing big shifts in gaming.
Don't let the annoying drivel and hype blind you to the genuinely useful possibilities.
Here are a list of AI use cases that I guess are "garbage" to you.
Detecting diseases. Creating drugs to cure or help with disease. Aiding astronomy. Understanding the genome. Scanning documents into text (OCR). Translation. Voice recognition. Detecting fraud. Spam filtering.
Are you willing to give up all of these? Given your attitude you probably should.
All of those are incredibly vague. I'm no astronomer, but I'd hazard a guess that it's "solving astronomy" about as well as it's "solving programming" or "solving mathematics". (That is, it isn't. Outside of a few grifters' imaginations, anyway.)
But sure, let's talk about a few of the more egregious ones.
> Scanning documents into text (OCR).
Useless for old texts, since you still need to review the transcriptions (and the "smarter" the transcription engine is, the harder review becomes since the errors look more plausible).
Useless for new texts, just type them in a readable format instead.
> Translation.
Heh.
> Voice recognition.
Useless. Do you really want more IVR menus?
> Detecting fraud.
Illegal unless you can explain your reasoning (tip: you can't, or you wouldn't be using "AI" in the first place).