|
My opinion (and I know nobody's asked for it):
"Smart" devices only extend technical problems to devices that should be simple. A TV's job should only be to display whatever content it's getting fed, end of story.
Since you can go to cbs.com on your PC/browser and play something from there, send that PC's output to your TV. If you can already do that with your phone, take that approach. Nobody's charging you any extra for whatever bandwidth gets needlessly (re-)transferred within your own LAN.
It's not ideal, but until these devices work as intended, you're only going to get frustrated trying to resolve those types of problems at that end.
|
|
|
|
|
|
Playing down problems is business as usual for bigger companies.
|
|
|
|
|
Interesting article and I get the issue...but it took the author a very long time to communicate the problem.
* AI egines (LLMs) have a reproducible bug that causes them to babble incessantly
* AI Companies either a) don't care or b) have no one with enough knowledge to understand that the bug could be severe (could cause AI to do something it shouldn't)
(or don't care and don't understand it)
This seems like the normal situation that I see from every company these days. 
|
|
|
|
|
They have a much larger problem than this bug - did they get permission to use published works to train their systems? I'm running into more authors who are putting disclaimers on the front of their eBooks that if the company want's to train using their work, they need to contact the author for approval and potentially payment. I'm waiting for companies like Amazon, with their huge Kindle Unlimited system, and library publishers to start filing copyright infringement cases against the LLM companies.
|
|
|
|
|
Except - of course - any LLM that Amazon decides to train on all those ebooks (especially the Kindle Unlimited). And many of those ebooks (especially Kindle Unlimited) are probably generated by LLM, so it's The Circle of Life...
TTFN - Kent
|
|
|
|
|
Quote: ...train on all those ebooks ... generated by LLM...
Reminds me of the old "Wiffenpoof Operation."
The Wiffenpoof is a mythical bird that flies in ever decreasing circles until it flies up its own a$$ and disappears.
Which, oddly enough, sounds like the phenomenon described in the cited article.
Cheers,
Mike Fidler
"I intend to live forever - so far, so good." Steven Wright
"I almost had a psychic girlfriend but she left me before we met." Also Steven Wright
"I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
|
|
|
|
|
obermd wrote: authors who are putting disclaimers on the front of their eBooks that if the company want's to train using their work, they need to contact the author for approval and potentially payment
I'm betting the net result of that will be that the LLMs start putting disclaimers in their responses to say that if anyone wants to use the response to train an LLM, they should contact the company for approval and potentially payment.
Because of course the "AI" companies aren't going to start reading and respecting those disclaimers; they're just going to point their scraping tools at the book, and blindly shovel it all into their models.
"These people looked deep within my soul and assigned me a number based on the order in which I joined."
- Homer
|
|
|
|
|
I'll play - what's an llm?
Charlie Gilley
“They who can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety.” BF, 1759
Has never been more appropriate.
|
|
|
|
|
ChatGPT is an LLM
M.D.V. 
If something has a solution... Why do we have to worry about?. If it has no solution... For what reason do we have to worry about?
Help me to understand what I'm saying, and I'll explain it better to you
Rating helpful answers is nice, but saying thanks can be even nicer.
|
|
|
|
|
Large Language Model - basically what's getting called AI these days (ChatGPT, Copilot, etc.)
TTFN - Kent
|
|
|
|
|
Perhaps Limited Learning Model is a more appropriate term, upon reading the article.
Or also Lagging behind in Leap towards Maturity.
modified 18hrs ago.
|
|
|
|
|
|
Exactly!
One of the problems I see is that the weakest link in the chat software chain, is context apprehension and continuity across probes.
To be fair, that's common amongst humans also. I'm 67 and I work with colleagues that are, on average early 30-somethings. They (almost, to be fair) never get my 60's, 70's and 80's cultural references.
I don't fault them for that, however. Culturally speaking, my internal LLM has been train on datasets they've never seen.
This is the common look I get: 
Cheers,
Mike Fidler
"I intend to live forever - so far, so good." Steven Wright
"I almost had a psychic girlfriend but she left me before we met." Also Steven Wright
"I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
|
|
|
|
|
Quote: Culturally speaking, my internal LLM has been train on datasets they've never seen.
 Good point ...
BR
|
|
|
|
|
What I read into this:
If you order a LLM to babble, it will babble. Took them a long time to say nothing.
Nothing to see here...
|
|
|
|
|
I suspect the source of the LLM problem described - setting aside all the "no one cares" and "I don't have a conduit to speak with anyone who might care" issues - is architectural.
To make an obvious "well DUH!" statement, there are multiple components here, just in the chatbots simple Input-Process-Output structure. The issue the article author talks about could originate in any one or more of them.
To me a bicameral or multicameral approach/architecture is needed. If these chatbot systems had an output monitoring AI that could "learn" to detect both garbage in and/or garbage out, that could help mitigate the issue.
I'm just flying by the seat of my pants here. I am fully prepared to be wrong, and if that's the case, please be gentle. 
Cheers,
Mike Fidler
"I intend to live forever - so far, so good." Steven Wright
"I almost had a psychic girlfriend but she left me before we met." Also Steven Wright
"I'm addicted to placebos. I could quit, but it wouldn't matter." Steven Wright yet again.
|
|
|
|
|
And I'm laughing way harder than I have any right to.
your code - YouTube[^]
Check out my IoT graphics library here:
https://honeythecodewitch.com/gfx
And my IoT UI/User Experience library here:
https://honeythecodewitch.com/uix
|
|
|
|
|
"A little time, a little trouble, your better day"
Badfinger
|
|
|
|
|
I'm still waiting for some friend to send me a little MP5.
GCS/GE d--(d) s-/+ a C+++ U+++ P-- L+@ E-- W+++ N+ o+ K- w+++ O? M-- V? PS+ PE Y+ PGP t+ 5? X R+++ tv-- b+(+++) DI+++ D++ G e++ h--- r+++ y+++* Weapons extension: ma- k++ F+2 X
The shortest horror story: On Error Resume Next
|
|
|
|
|
I'm not picky a regular size would be just as easy!
|
|
|
|
|
And I hope the renessance will come in my lifetime...
"If builders built buildings the way programmers wrote programs, then the first woodpecker that came along would destroy civilization." ― Gerald Weinberg
|
|
|
|
|
build the game or build.com?
"A little time, a little trouble, your better day"
Badfinger
|
|
|
|
|
I use DuckDuckGo exclusively and it is down right now (since 4:21am Eastern) -- see image of their tweet[^] or see the actual tweet at: x.com[^]
I also just tried to do an Image search from Edge and got this error from Bing[^].
What's up with Search Engines?
modified yesterday.
|
|
|
|
|
raddevus wrote: What's up with Search Engines?
In general terms? Too many ads, too much AI-generated BS, and too much keyword-stuffing SEO-optimised crap filling the first page(s) of results for them to be much use any more.
At least DDG doesn't suffer too much from the first two.
"These people looked deep within my soul and assigned me a number based on the order in which I joined."
- Homer
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
