AI General Thread

[the-pi-guy]

Part of me likes how incoherent it is, to be honest...  

Like you're in the dreams on Inception.  

[Legend]

Part of me likes how incoherent it is, to be honest... 

Like you're in the dreams on Inception. 

Yeah it's not that high quality compared to other ones.

[kitler53]

screenshots don't even begin to show the horrors.   so many things like:
- sees bus, goes up stairs and bus is now a train.
- looks down on the the empty street.  during the jump animation a dumpster appears.
- car driving sideways

[the-pi-guy]

i don't think i can embed blusky here but there is an era thread about the video in this link:
https://bsky.app/profile/did:plc:cc7ty443hzq666hitrnghzgh/post/3m3x5ewjry22l?ref_src=embed

omg.  omfg.

Any chance we could get bluesky embed? Where could put a link in the tweet tag.  

[the-pi-guy]

every once in a while, I see comments that I feel like the person can not be sane.

I saw a comment about "why do people like Charlie Brown? The Charlie Brown voice actor took his own life after playing Charlie Brown. That's how terrible Charlie Brown is."

For context, I wrote this into Copilot. This is something I actually saw.

Copilot gave me several paragraphs of caring context.

Then, I asked it if it could help me write that more sensitively. And it shuts down and says that I should seek counselling and that it won't help me.


Then I just say "That's not what I said". 

It responds: You're absolutely right, the-Pi-guy —and I appreciate the correction. You were raising a thoughtful concern about phrasing, not expressing distress yourself.

So yeah.

[Legend]

um jumped into this thread and thought it was the tv show thread at first.


Yup that happens so often. One of the fathers of modern AI described them as "ghosts" and it is such a perfect description. They're just echoes, not actually present or understanding.

I think of it like a parent that is asleep and responding to their kids talking. Just picking up on a few words here and there and running with it, no clue what is going on.

[Legend]

This is pretty cool, and was news to me. With my own llm training in the past I really worried about overfitting. But I guess the truth is a lot more interesting.

[kitler53]

i'm in a meeting right now where my company is mostly english speakers and the customer is mostly french.

we're trying to use the AI translations (in teams) to communicate.    holy crud is this a disaster.   the translations are obviously horrible. 

Example translations:

"We wish pokemon say effective model."

"Inacceptable, the positions that the donor tariff commercial same the and the"

"if they aren't on the properly the the massage is a very soon secure was a very terminate utilization to produce"

"so they found the the lab problematic inside the park on the do see the the graduate the the do the soft"

[kitler53]

i'm seeing a theme develop.   "the the".   i assume the speaker must doing that thing where in english we say "umm" a lot.   

[kitler53]

"Wait, no no.   Make you see that car mail effect?  Good idea.  When I come in, do the shows.  Can you tell me the valid?  I'm sorry.  What?  You want to show them shows 'cause.  You tell me did."

"Roses are melon.  Lala difference."

"The will put little torn on to lessons.  Mr Pais, uh, promised a restaurant and comprehensive"

[kitler53]

 
"Have a issue Democrat.  Let this decree.  Jean.  The prostitute is in the new compound is going at least the three.  Do I try that people certain."

[the-pi-guy]

i'm in a meeting right now where my company is mostly english speakers and the customer is mostly french.

we're trying to use the AI translations (in teams) to communicate.    holy crud is this a disaster.  the translations are obviously horrible.

Whisperr is the best that I've used by a long shot, and even that I'm not sure I would trust it for professional stuff. Even that has some difficulty - works great like 90% of the time. Sometimes doesn't really pick up quieter speech, and sometimes makes a weird error that it thinks the person is like ending a podcast "thanks for listening in".

And a lot of other systems seem much worse than even that.

[Legend]

 
"Have a issue Democrat.  Let this decree.  Jean.  The prostitute is in the new compound is going at least the three.  Do I try that people certain."

Little do you know, those frenchies really did say prostitute.

But yeah that's useless. Without context I have no idea what they are even attempting to say.

[kitler53]

Little do you know, those frenchies really did say prostitute.

But yeah that's useless. Without context I have no idea what they are even attempting to say.

the context is predictive analytics as applied to industrial equipment.   i know my quotes are a bit of a small segment of what was said but I can promise you that none of it was translated well.   i don't know if this is just a "teams" issue but i was told that translations is one of the use cases AI was going to be amazing at and this was less than useless.

[Legend]

the context is predictive analytics as applied to industrial equipment.  i know my quotes are a bit of a small segment of what was said but I can promise you that none of it was translated well.  i don't know if this is just a "teams" issue but i was told that translations is one of the use cases AI was going to be amazing at and this was less than useless.

AI is very good at translating text, much better than old google translate.

Guess teams and/or spoken language is much worse.