logoalt Hacker News

univocalyesterday at 9:13 PM3 repliesview on HN

You know what's even worse? That if he had tried any of those "here look! we can write it for you!" tools he’d have found out that they don't even work.

Gmail summaries are nonsensical most of the times. The suggested replies completely miss the intent of the original message I was trying to write.

Most AI integrations around are basically alpha-quality code, that if there wasn't this forced pressure to adopt AI, AI, AI at any cost, they wouldn't have been shipped in this state at all.


Replies

serial_devyesterday at 9:30 PM

Oh I got an email about a booked flight, Europe, Denver, Vegas. For some reason the times weren’t picked up in my calendar so I naively thought I’ll try their AI tools to put it there…

I tried to get it to work for five minutes, it couldn’t get it to work.

Then I was so pissed that tried for another thirty minutes to “prompt” my way to get the events created correctly, highlighted the timezone issues…

Then gave up and did it manually in 2 minutes.

show 1 reply
Tor3today at 12:54 PM

Research on LLMs show that they so far can't summarize. They fake it by just extracting stuff, i.e. making the text shorter. They, as you found, can't understand intent. Or what's actually important. I never read summaries generated by AIs, it's a waste of time, and worse.

agiacaloneyesterday at 10:02 PM

Because the whole goal is not to help us.

It’s to train their AI models. You hate it and then fix it. AI gets “better”.