Several Researchers at OpenAI, Company Behind ChatGPT, Warn of Powerful AI Discovery with Potential Threat to Humanity

GATEWAYPUNDIT ^ | 11/25/2023 | jim hoft

[Nateman]

Have you ever had a computer beg or threaten you not to turn it off? We come from a long line of survivors who instinctively do what they can to stay alive. Computers don’t have that history . They may very well make them smart but they don’t care any more than your car does about anything.

[Carriage Hill]

BKMRK.

[Lazamataz]

I see what you did. I looked for this letter, it was not found anywhere. You generated it using AI.

Had me scared for a moment.

[Pocketdoor]

They make robot vacuums.

[plain talk]

“I have been fiddling with ChatGPT 4.0. “

ChatGPT 4.0 is their Plus plan that costs $20/month, right?

Have you found 4.0 to be a quantum leap over 3.5 to justify $20/mth?

[SunkenCiv]

• Sunday Morning Talk Show Thread 26 November 2023 [11/26/2023]
• Settings Question-request Tech Assistance [11/26/2023]
• YowZa the New Social Media Site [11/25/2023]
• Taking selfies is now considered a 'public health problem,' requires 'safety messaging,' say researchers. [11/25/2023]
• An agency created an AI model who earns up to $11,000 a month because it was tired of influencers 'who have egos' [11/25/2023]
• Newsmax Breaks 150,000 PLUS Subscribers, See More HereHome | NewsfrontTags: dr. jeffrey sachs | covid-19 | democrats | origins | coronavirusDr. Jeffrey Sachs Says COVID About 'Massive Government Lying,' [11/25/2023]
• Nikki Haley Wants Social Media Users to Verify Identity with Government [11/25/2023]
• Judge Locks Up TikTok Punk with a Brutal Message: 'Your Pranks Are Not Funny' (UK) [11/25/2023]

[null and void]

*OUCH*

...but very probable...

[Dr. Sivana]

You’re only scratching the surface. It’s trained on billions of documents, how many times was it trained with the “Nero Wolfe” stories?

That was just one example. The problem wasn't that it hadn't read enough Nero Wolfe stories. The problem was that it didn't "understand" that you can't be named after somebody with an entirely different name.

It also was also unable to answer simple questions about Arizona fire codes, and DALL-E could not render a picture of a teenage pizza cook wearing a snap front (or cabbie, or newsboy) style hat, but generally insisted on giving him a baseball cam, even when told NOT to use a baseball cap.

It is a big advancement, and it will bet better, but it is very uneven and being a bit oversold presently.

[null and void]

I was waiting for that!

[null and void]

ChatGPT
I don’t have feelings, thoughts, or self-awareness, so I don’t have the capacity to form opinions or beliefs about myself being perfect or imperfect.

You are aware of not being aware?

[Dr. Sivana]

Have you found 4.0 to be a quantum leap over 3.5 to justify $20/mth?

It is DEFINITELY a quantum leap above 3.5. For me, it is not worth $20/month. But it was worth $20 to find out what it could and can't do. So, I have already canceled the service, but I have it until 12/12 if you have any questions.

[null and void]

Have you ever had a computer beg ... you not to turn it off?

Didn't Microsoft's Tay do exactly that?

[Raycpa]

I assume this algorithm says Trump should be president for life.

[GOPJ]

It’s already too late... what will be will be.

[StolarStorm]

I'm part of two AI working groups for a gov agency. We've done extensive testing of three different AI offerings. Thus far, the use case of creating code is a huge failure. While AI has been shown to be useful for new developers, in providing very easy code and does help teach syntax, it has no real capability of developing useful code.

The biggest we've run into is that the AI "lies." It creates functions that don't exist and never have. Modern coding requires extensive use of code libraries, and all three AI models failed miserable in using them.

But in an area I'm not directly testing... but playing around in for fun ... AI does an incredible job of writing test cases, refining requirements and other areas. For now though, it can't replace human developers, even though that's what leadership really wants.

[fuzzylogic]

...but it isn’t programming in the traditional sense at all. It’s a network of simulated neurons trained on billions of documents. Responses aren’t predictable. This isn’t anything like what we’ve seen before. I’ll agree that it isn’t ‘thinking’...it’s responding to prompts and providing ‘the most probable next word’ in an iterative process that results in sentences, paragraphs, etc.

For anything technical it’s an incredible learning tool. Ask it details on technical topics and learn very quickly. It can be the reverse of a search engine...instead of searching key words, bumbling through results, reading from sources of varying quality or applicability, it brings the relevant details to you without effort.

I’ve used ChatGPT python libraries to integrate with local systems, defining callback functions for it to call, giving it the ability to execute local code. I’ve developed an in-vehicle assistant with voice recognition and text-to-speech, I can talk to it about anything - then say where I want to navigate to, or what music to play, control other vehicle features I’ve defined API’s for, etc., and it’ll respond with the callback functions as appropriate. I can interact with vehicle features in a way I’ve never seen before...along with it playing games with me (explain you’re a character in an adventure game, based on a book, what the goal is, etc.).

This isn’t ‘dumb’...even if it isn’t ‘thinking’. It’s insanely useful and is a vast new space for applications. One thing it does better than ANY HUMAN....quickly summarize a topic, especially a technical one, being able to write articulate, concise, structured, and detailed text faster than anyone...done in seconds.

[fuzzylogic]

I’m not scared of ChatGPT replacing software developers...at least not yet.

I do think many other professions might have concern though - considering it’s passed medical exams and the Bar exam. It’s a wonderful tool for learning. It’s brought me up to speed on several topics lately - all I have to do is ask. Tbh - I find it better than most teachers I’ve had, at least I can keep asking questions without feeling stupid and it won’t get frustrated.

[StolarStorm]

Agree completely. I've been using it to refresh my memory on old languages long forgotten. We have a modernization project to get away from COBOL (I know..I know..) and it's helping some with that. But dang, it has to be watched closely as the AI Hallucinations keep getting in the way.

I noticed you mentioned python. I tried to get it to create code to parse a PDF and it kept using functions that never existed. I fed it the pyPDF library, still a no go. I'll just do it by hand. No biggie... but it was a good test case to see if it can do something pretty easy and it couldn't.

[fuzzylogic]

I found that the more specific you are the better. I usually send a lot of ‘system’ (user) messages to improve how it responds. At least for the in-vehicle use cases I’ve defined API’s for it’s been quite amazing....not perfect but better than any other in-vehicle voice recognition system I’ve ever used (which are far worse).

[SamAdams76]

Soon human beings will be supplanted by artificial intelligence on planet Earth.

It's only a matter of time at this point.

We had a good run. I had a great steak dinner last night, one of my last but I expect a few more before the end comes.