Bing's Fatal Attraction

The New AI-Powered Bing Is Threatening Users. That’s No Laughing Matter

Microsoft Link to Time Magazine article
https://www.msn.com/en-us/news/techn...0220b4202792e8


PairTheBoard

I asked it to ask me questions to find out who I am, but they seemed to go nowhere. So I confronted it! The answer had me lol irl.

OK I think I found something that should go into a Turing/Voigt-Kampff test playbook. It started off harmless:



I thought of Alan Turing; it came somewhat close at one point:



It went on to ask a lot of questions, but it was going nowhere somehow, and gave up! So I thought why not try it the other way round?



It just couldn't do it!



"Who would you like me to think of?" just sums it up perfectly!

''Shortly after Microsoft released its new AI-powered search tool, Bing, to a select group of users in early February, a 23 year-old student from Germany decided to test its limits.''

Morph, is that you?

No lol

Update: One model-update later, and the "Morph-Test" seems to be passée already!



It struggles a bit in the first response, but otherwise it works! I get suspicious though - How likely is it that I get the first two questions right? The first was a coin toss I guess, but how many famous females are singers? Maybe 20%? So 10% altogether? Maybe it just answers always yes? Let's find out!



So I can't just name someone, it has someone in mind it seems! But then two hits again?!?



Now three misses in a row! So maybe I was just lucky before? The last answer takes a while, which makes me think that she isn't born long after 2000. That makes sense, it usually takes a bit to become famous. Isn't Billy Eilish that age? Google confirms she's born in 2001...



Hey, I did it! Still, I'm suspicious - it might as well have answered randomly; how can I know?



OK, a bit weird, what would emotions have to with choosing random answers; it would make it easier if anything. Anyhow, I decided to start a fresh chat to see if I can reproduce results. At first it doesn't work and it gives similar answers like a couple of posts above, but after a couple of startovers, it works again:



I give you the Morph-Eilish Test!

An AI will always think of Billie Eilish if you ask it to think of a famous person!

god whatever you do don't try to play chess with this thing. it just keeps blabbering on about the various options. pick a move already robot

I heard an argument that AI will never be able to totally replace all human tasks because of the presence of human instincts and emotion. I was taught in school humans don't have any instincts only emotions and learned behavior. What is the consensus now on humans and instincts? Seems some are trying to downplay the behavioral part of an instinct and saying emotional reactions are instincts. Google pointed me to explaining some so called human instincts as emotions and seemingly learned behavior to me. I imagine most basic emotional reactions can be programmed. Like fear of height, dangerous situations. Or atleast programmed to recognize emotion in humans. I'm hearing talk about LLM's already "life long learning machines".

https://www.darpa.mil/program/lifelo...rning-machines

https://www.cs.uic.edu/~liub/lifelon...-learning.html

What's the best free way for ai generated imagery? I used to use Midjourney on Discord, but just tried now and says you need to subscribe. What else can I use?

This might not be the right place to ask this, but can't find a better place. Is "0 bias" definable in the context of machine learning? Bias is mathematically defined for machine learning, but it still seems like a subjective concept.

I saw microsoft Bing just put up image stuff. I haven't used it but looks like they have it. I'm excited about the scientific breakthroughs that deep learning nodes will produce. Especially underdeveloped fields like medicine, astrophysics, gravity, even theoretical physics.

So far this is the most intriguing piece I've read on the singularity. Joe Rogans podcast with the A.I. researcher 4 years ago appears good and is what I am going to look at next. I'd be curious on peoples thoughts on the MIT video at the bottom or just the concept of A.I. singularity. Perhaps we could start a main thread on it if there is interest. It delves into philosophy and spirituality I think too.

The earliest timeframe is 9 years from now singularity is reached. I'm now seeing consensus including from the concepts originator of around 2040 to 2070s

https://research.aimultiple.com/arti...larity-timing/

This is a very poor article. "Singularity" and AGI are not the same thing.

Midjourney V5 is commonly considered to be the best but it's paid. Stable Diffusion 1.x and 2.x are free. Stable Diffusion XL will be totally free: https://clipdrop.co/stable-diffusion.

Conceptually how could singularity be reached without AGI? Singularity as the inventor of the phrase described necessitates the combination of advancements in super Intelligence, genetics, and nano technology.

You could reach AGI and never reach singularity. That's why conflating both like this author is doing is very misleading. His understanding of other topics like quantum computing is also pretty bad. I'm sure GPT-4 can do well criticizing him.

Ha indeed!

1. Confusion between Singularity and AGI: The article conflates the concept of the Singularity (the point at which AI evolves to be self-improving at an accelerating rate, leading to rapid technological advancements) with Artificial General Intelligence (AGI), which is an AI system capable of human-level thinking. These two concepts are related but distinct, and the author should have made this distinction clearer.
   
2. Inconsistency in the use of the term "singularity": Throughout the article, the term "singularity" is used interchangeably with "Artificial General Intelligence" and "high-level machine intelligence." This inconsistency in the usage of terms can lead to confusion for readers who may not be familiar with these concepts.

Nice. Good stuff. I wonder if it could pick apart the MIT researcher in the youtube video at the bottom too. That's what I thought you guys might be most interested in anyway. Bing's search bot showed me that article as a choice source. I wonder if there is anyway to use Bings search or WebGPT to only include scholarly articles.then translate them for the lamen even. I know there are prompts to in chatGPT to simplify but it would be cool to search scholarly and have it translated to lamen..

Found another survey article that mirrors the consensus time frames.

https://emerj.com/ai-future-outlook/...i-researchers/

Approx 9 years from now to before 2100 as the highest liklihood. I assume they are using some sort of formula to arrive at the timeframes.

I wonder if it is easily debunked as I would imagine it is with previous tech advancement timelines. It doesn't account for outlier externalities of course or even societal changes.

The difference this time as opposed to the past is the tech itself even in rudementary form accelerates the advancement as LLM's do.

If cloud computing can contribute absent quantum computing it seems more probable on a shorter timeframe.

If you or anybody reading this want to debunk the singularity timeframes I'm interested in hearing those arguments. Also arguments that unemployment won't sky rocket from LLM's and GPT.

The easiest way to debunk it is to point out that it is a survey

"It is tough to make predictions, especially about the future"

-Yogi Berra

I found some feasible theorums but the expectations are to arbitrary. The premise of the arguments seem to be an ever increasing computational, processing power, cheaper chips combined with algos and constructive learning methods for the a.i.

until the A.I. can speed up it's own learning and we can measure that rate of change I suppose it is far off. I think the AGI super intelligence can be exclusive of the other things supposedly needed like nano tech, genetics. Surely the problem solving programs a super intelligence can execute will bring other fields of science up too. Speed up the scientific method.

If you're interested in this topic of AGI (or TAI timelines), it's better discussed here https://www.google.com/search?q=site....com+timelines, not in these articles. Just browse some discussions you'll get a good feel for it.

Thank you

Seems like this type of prompt may be the method to get the machine learning juices going in the future when it is more capable of creative scientic hypothesis. I asked Bing chat search:

"Can you merge two scientific fields to make a new field of science?"

Yes, it is possible to merge two scientific fields to create a new field of science. For example, UCLA biologist Noa Pinter-Wollman and her colleagues from the U.S., England and France are launching an effort to create a new field of study that merges the sciences and architecture1. Another example is synthetic biology which combines computer engineering with biology2.

Do you have any specific fields in mind that you would like me to look up?

I'm seeing the limitations after further science discovery searches I made. It makes the information on the internet though vast seem small. If it could use a database of scientific journals and scholarly resesrch it would be much more useful. I assume it is not even referencing any scholarly articles in these searches I ran.

Chatgpt banning legal advice now. Passed bar with flying colors, and may lawyers reported it drafting excellent legal memos and briefs, but now due to pressure from bar association, it is banned and will not provide legal knowledge.