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

GATEWAYPUNDIT ^ | 11/25/2023 | jim hoft

[bitt]

Recent findings at OpenAI, the Artificial Intelligence powerhouse and creator of ChatGPT, have incited an internal alarm just as the company’s CEO, Sam Altman, faced a brief but compulsory retreat from his position.

Days before a whirlwind of corporate upheaval, several of the firm’s researchers reportedly penned a concerning letter to its board of directors. They highlighted a significant AI breakthrough with ominous implications for mankind, sources with insider knowledge told Reuters.

A confidential letter, signed by several staff researchers to the OpenAI board of directors, highlights concerns regarding a powerful artificial intelligence feature or algorithm. The letter, not made public, played a crucial role in the events leading up to Altman’s removal from his position.

Sources indicate that the board’s decision was influenced by a range of factors, including concerns over the premature commercialization of advanced AI technologies without fully grasping their potential consequences.

In the tumultuous days leading up to Altman’s firing and subsequent return late Tuesday, a wave of unrest swept through OpenAI. More than 700 employees reportedly threatened to resign, expressing solidarity with Altman and considering a move to Microsoft, a major backer of OpenAI.

In response to inquiries from Reuters, OpenAI, while declining direct comment, acknowledged in an internal message the existence of a project referred to as Q* and the letter to the board. The message, disseminated by Mira Murati, a senior executive at OpenAI, seemed to brace staff for upcoming media stories, without confirming their specifics.

[fuzzylogic]

Also...I think the biggest problem with writing code, you can rarely, if ever, define the requirements to be specific enough. It’s great for some educational boilerplate to code (e.g. write me a based client/server socket comms code in C) but for anything more complex, requirements are usually so poor, even for humans, that to expect it to write anything large or of real substance is asking too much.

[StolarStorm]

Sounds about right. Just like offshoring code (ugh), you have to be very precise in writing requirements (prompts). Humans (other than offshored coders) have the advantage of subject matter expertise and intuitive thinking. Ok, some humans, not all. lol.

[rod5591]

“Firstly, we must stress the importance of maintaining complete confidentiality throughout our communication. The nature of this correspondence necessitates utmost discretion to avoid compromising the integrity of our research and development processes.”

LOL, we are reading the “completely confidential” memo on the Free Republic internet forum! Can’t make this stuff up!

[adorno]

...but it isn’t programming in the traditional sense at all. It’s a network of simulated neurons trained on billions of documents.

Not traditional programming? But, still it's programming. Meaning that, it' programmed with a lot more data/information available to it. With more data or information available, of course it's going to come up with a lot more 'useful' information than any system before. In the future, a lot more useful info will become available, thereby making the systems seem 'more intelligent'. In essence, the more info, the more things seem to be intelligently produced.

It's about the intelligence, stupid. (And I'm not calling you stupid, btw.). Intelligence is still about a lot more than data or information. Data or information cannot produce 'intelligent results', without the intelligence, and the intelligence is still what's trying to emulate what humans do, but they're still way off in that department. For example, can an AI app look at the heavens and 'wonder' how things got that way, and why, and by whom?. Only a human can try to absorb what's available in his/her environment and try to figure it all out. A computer with AI has to be fed that environmental information, and then it will be stuck with doing nothing about it all. A human comes equipped with the intelligence to do something about what he sees or feels and hears and touches and smells. AI has to be told the information, but can't react on its own with more intelligent questions and answers.

Human are still trying to do what they've done with computers for ages, and the main difference today is the vast amount of information that's available and how it's organized. In essence, AI organizes the information a bit more intelligently than in the past, but it's still not 'intelligence'.

[StAnDeliver]

This Q(uantum)GPT is a coder-in-your-pocket, and the freakout is over the disruption.

GPT-4 itself is capable of elucidating/extracting more valid code than I ever thought possible; and yet Sam gave an interview nearly a year ago where he was already talking about GPT-6; and reading between the lines, the takeaway was (and is now ever moreso):

• AGI-positive is happening in this decade, not 2032.

• The first labor group to be seriously detrimentally-impacted, unironically, is probably coders, or at least 'junior devs' as even a thing, along with all marginal overseas outsourced rent-a-coders.

• If you thread the commonality between a BA and a Dev, it was always logic, and since AI is exquisitely logic-based, Q(uantum)GPT blurs those lines by offering both a pathway to the desired outcome, based on their ability to precisely extract it querying Q(uantum)GPT, and hopefully recognize any hallucinations before committing it and passing it along to QA. Agile looks a lot different once the next GPT hits the fAIn.

• In this scenario, a BA or PM who migrated from Dev and can wrap their head around AI queries to satisfy Business demands, will now drive Agile, until one B2B emerges to own the field on simplified front-end AI querying.

[fuzzylogic]

Allowing businesses to have more effective ways of querying data is going to explode - of this I have no doubt.

That is very different than the system design and the detailed requirements necessary to design it. All of the systems I’ve ever designed never had clear requirements, the main challenge is to get them defined, at least to the point where you can actually design. Even safety critical systems with stringent requirement management processes still have requirements that aren’t as concise as they probably should be.

I’m also not sure we should try and thin the pool of junior developers even if we can. Long term we need experienced developers, I don’t see that ever changing...unless these AI technologies become so superior that we’re basically handing the keys over to them, which will also be way beyond writing code. This seems like a bad idea :)

We’re already seeing code assistants in Visual Studio, this will continue. I’d be far more concerned if my profession revolved around ‘standard knowledge’, not new design. This would include teaching, medical ‘standard’ care, the law, etc.. A lot of this information is static and unchanging. They just require somebody years to learn it all.

[Fledermaus]

Yep. It’s still humans that program the machines.

[Fledermaus]

Yeah, planes were going to fall out of the sky . I guess physics wouldn’t work.

[Fledermaus]

Spock would destroy AI with Vulcan logic.

He took out Norman in “I Mudd”.

[Right Brigade]

“That is the truth, but they are not selling the truth, they are selling a lie to manipulate.”

Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.

Frank Herbert, Dune

[usconservative]

me too, and when I scrolled through the entire thread and no on else posted it yet, I was shocked. LOL! 61 year old (as of tomorrow) Star Trek nerd.

[daniel1212]

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.

See

[Video transcript] Can Scientists Answer These Questions? RNA, Abiogenesis, Chemical Natural Selection & more Dr. James Tour. [More of his videos here]. Dr. James Tour ^ | 8-24-2023 | Dr. James Tour

I used https://poe.com to format this, and to which I only bolded section tittles. I could not paste too much at once (about 2,000 words), and sometimes I needed to to repeat the instructions. And which were,

“I am going to paste sections of a transcript (which is a “wall of text” having no paragraph spacing or punctuation) and I want you to render each section of a transcription readable (to humans) by only adding punctuation, and paragraph spacing, and capitalization of pronouns and proper names. So here is the section to do that to:“

[StAnDeliver]

"I’m also not sure we should try and thin the pool of junior developers even if we can."

Oh, aboslutely. Still, I think it's going to be a weird time for emerging devs ... they will be expected to have mastered tip-of-the-spear AI as the primary code-adjunct tool, and at the same time, not piss off the old hedgehog senior dev manager -- that just hired them -- who is hanging on because of his 25 years of institutional memory and has limited interest in AI.