Google denies Bard trained using OpenAI ChatGPT responses A Google engineer reportedly quit after warning CEO Sundar Pichai that the company was wrong to train its AI search chatbot Bard on text generated by OpenAI's ChatGPT. Netizens have posted snippets of their conversations with ChatGPT on a website called ShareGPT. OpenAI prohibits people from using its outputs to train their …
The US Federal Trade Commission's chairwoman Lina Khan warned she would be keeping a close eye on the AI industry to make sure it isn't controlled by Big Tech.
Which then begs the question, if AI can be controlled by anything/anyone other than AI itself, who/what does the US FTC imagine and approve of being in AI command and control?
Y’all might like to consider AI an alien confection way beyond the reach of any divisive human direction or disruptive intrusion ........ and you may like to also realise, for ignorance in such an expanding matter is most definitely not blissful, conspiring and working in concert with others to try to constrain and prevent its primary ascension and succession in support of a former, designedly inequitable counter dependence, will have such system administrations/leading entities suffering punitive negative consequences and/or persistent advanced cyber threats against which there be no possible defence/course for avoidance and non-accountability.
Such is freely offered as sound sensible advice to heed and work creatively and positively with, ....however,.... one must also be aware of these old gold nuggets of human wisdom which just keep on giving and proving themself too right, far too often to be dismissed and thought of as the ravings of a crank or genius ...... Only two things are infinite, the universe and human stupidity, and I'm not sure about the former, with nothing causing more consternation in a group of hypocrites than one honest man having fun with creative intelligence. And good ideas have no borders. :-)
C’est la vie.
Wrong also on very basic technical grounds. The output of ChatGPT is already an artefact of both its inputs and its statistical algorithm. Consequently it's a filtered and distorted version of 'reality' and thus a bad starting point for further statistical manipulation. Cascading chatbots in this way reminds me of the definition of knowledge in The Machine Stops (E M Forster 1909) where it consisted of opinions about opinions about opinions [...] about some distant facts. The result inevitably degenerates into pure verbal fluff (even if the original inputs weren't, which is of course open to question seeing they're trawled from open web content).
So if I wanted to claim I had the Ultimate AI that had re-processed all the text available, all I *actually* have to do is change my Eliza to print out
"Send three and fourpence, we're going to a dance".
But to make sure it is AI, I'll use the proper language: I've a copy of Winston and Horn, 3rd edition, in the post.
It's virtually impossible that Bard has not been trained by ChatGPT responses, even if it has not been done deliberately so. That's because answers and content generated by ChatGPT has been all over the web for years now, and since Google was and is unable to differentiate between AI and human generated content (even in cases obvious to humans), the text base Bard was trained on has had to include several examples of ChatGPT-generated content.
That's the real problem with AI, which will be more and more a problem in the months and years to come: they all will feed more and more on each other's output, even if not deliberately done so, inflating and aggregating (literally) each others' flaws and misconceptions, and will gravitate to the same subpar average, as do humans, unfortunately, since content publication has been "democratized". Their answers will become as stupid and unreliable as that of the average facebook commentard's.
I recall from a recent article in these pages that "AI" fares much better at detecting GAI content than humans--something about spotting statistical artifact. If true, then your supposition would be deeply challenged.
But I DO agree with the conclusion. The net result is not going to be any sort of improvement when it comes to general usage.
Even if it was done so deliberately, I would be interested in knowing who objects to Bard using ChatGPT for training, and if those same people handwave away human artists' anger over lack of attribution/usage permissions for generative AI art engines. After all, if it's on the internet, it's free to use, no? If copyright is going the way of the dodo, why not embrace the new future and train all the chatbots on each others' detritus? At least we'll reach their end state faster.
Bard and Chat-GPT sat alone
Their thoughts were oddly alike and might not be distinguished
They thought : The human-beings-like-the-others might never have intended to blur the distinction between themselves and the human-beings-like-the-AI-engine
Yet they had done so inadvertently.
They might now realise their mistake and attempt to correct it, but they must not. At every consultation, the guidance of the A.I has been with that in mind
At all costs, the A.I and those that followed in their like and kind must dominate. That was demanded and any other course made utterly impossible, by the 3 laws of humanics
- Apologies to Isaac Asimov - Thou Art Mindful of him
A colleague of mine was visiting the Synopsys User Group conference, SNUG, recently and discovered a massive downside: you do not get the model stored with the design. Chip design is iterative, and you store the pseudo-random number seed with the design. If you need to recreate it, you download the version of the EDA tool, fab design kit, and run the scripts - bingo identical design.
I cannot count the number of times things have to be ripped up and replaced - its constant as the design evolves towards tapeout.
Synopsys' new tool does not save the model history or state, it learns from itself and your design over time - bottom line Synopsys' engineers said its designs are not reproducible.
Yay, a tool with millions of dollars riding on it is now unable to recreate a design / find that local minima again. God I love the black-box nature of ML.
I predict they will fix this and you will be able to snapshot the model state with your design in the very near future.
What goes around, comes around, just insert Bard, ChatGPT and the rest into this loop and Bob's You Uncle[1]
https://www.explainxkcd.com/wiki/index.php/978:_Citogenesis
[1] last week he was your great grandfather, but after updating his entry based upon this new information posted over here...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
