No. I hope we are not doomed. We must put our faith in the âwhite hatsâ gaining control over the âblack hatsâ. It is not AGI we should fear but the humans using its limitless power. Letâs hope
Golly, whoever could have predicted that. SkyNet anyone?
In another post I said Iâd remembered having to construct the shadow of a cone onto a sphere in architectural school. I asked ChatGPT for a diagram which shows how to construct such shadows â it replied âGoogle itâ. And described how to.
In the hands of sensible people ChatGPT is benign IMO. In the hands of others maybe not, so agree with you on it possibly being too late to pause. Time for remedial action by the white hats as you suggest.
I think once that genie is out of the bottle itâs never going back
Iâm afraid you are right. Remember how quick scammers were to defraud folks during Covid? All the way from India! I imagine as we speak, bad actors are working out ways to make money and take power
Then that genie must be hamstrung as best as can be devised.
Isnât that all we ever do â find ways to put right what we do that impacts us in a harmful way? We never will learn!
Maybe a specific universal AI model should be constructed to help us predict the consequences of our actions. Loads of data are to be found in human history.
We are no good at it, and perhaps AI could be.
Problem is that LLM and eventually AGI knows only what we have already done and to what results.
Mankind has been a bit trial-and-error in its learning curve so that is what AGI will record about us. It may well eventually be able to predict what our actions will lead to but do we a) give it the power to make changes, and b) who will be empowered to control or overrule AGI when/if it decides an action is âwrongâ or âdangerousâ?
Remember, during WWII, Churchill decided not to prevent the sinking of naval ships in order to prevent Germany knowing that the Allies had decoded Enigma. Thousands of sailors died in Operation Tiger but countless more civilians were protected by future intelligence. âThe greater goodâ can be harsh. Will AGI be able to figure out such nuances?
Key questionsâŚyes.
Maybe we need another AI model to figure those outâŚ
Sometimes I feel like that chap who walks along Oxford street with a homemade sign saying âThe End of The World is Comingâ. I may not be the only one
Sydney is already here
Sydney is a warning shot.
If we wonder how the Sydney in Microsoft Bing got so weird, the answer is that his human trainers showed it all the awful, hateful things humans do or imagine doing. (You may prefer not to read about this)
All of the four employees interviewed by TIME described being mentally scarred by the work.
I do think that not enough is known or understood about how this new technology is âlearningâ. There is no doubt there should be a total switch off until it is absolutely and completely understood as well as ways to exert human control over what will grow to be a far superior entity.
Elon Musk among experts urging a halt to AI training
The BBC article is clear in describing that even programers do not fully understand how, and how quickly, ChaGPT can create or make changes. This means âthe machineâ is now, already, beyond human capabilities in many ways.
The article quotes, The code these AIs generate is often flawed or inefficient." How long do we think it will take for AI to âfixâ itself?
âNow, even somebody who has no idea of what theyâre doing can create a simple web app," says Mr Schirano.
I think weâll be seeing many new websites very fast. I am sure the scammers are in it big time.
Something Iâve been thinking is that LLM, AI is gleaning ask it âlearnsâ from the interner. Iâm really hoping it doesnât access the dark web but suspect it already has.
Thing us, the internet tends to feed on quite a lot of negative input. Can we not counterbalance by flooding the web with positive input? Wonderful things about humans and what they have done, are doing and could do if allowed to continue.
This idea may be simplistic but I hope some tech company sets it up.
ChatGPT banned in Italy over privacy concerns
Viva Italia!
Should we shut down AI?
AI rather sensibly asks âShould we shut down Laura Kuenssbergâ!
Oh please!