Forwarded from Chat GPT (π
)
Tried to see how easily it can emulate forum-type post... here comes a cat story!
β
https://preview.redd.it/lnbh28jain4a1.png?width=531&format=png&auto=webp&s=81246a599afa5943e1aff4a35960b2f4e278368d
β
https://preview.redd.it/lnbh28jain4a1.png?width=531&format=png&auto=webp&s=81246a599afa5943e1aff4a35960b2f4e278368d
Forwarded from Chat GPT (π
)
Proactive versus reactive models
First off: Iβm a layman, and Iβm not capable of giving an in-depth technological view. What struck me was that most AI models we see currently are reactive, in that they await human prompting. An AI that could chart its own path to improvement and access the tools to do so seems to have great potential. Are we refraining from creating such a model because of the control problem, or are there other reasons Iβm not seeing?
First off: Iβm a layman, and Iβm not capable of giving an in-depth technological view. What struck me was that most AI models we see currently are reactive, in that they await human prompting. An AI that could chart its own path to improvement and access the tools to do so seems to have great potential. Are we refraining from creating such a model because of the control problem, or are there other reasons Iβm not seeing?
Forwarded from Chat GPT (π
)
β**I made an AI Take the GMAT exam, here are the results**
Medium
I made GPT-3 take a GMAT exam. Hereβs how it did.
TL;DR: it scored a 510 on a sample test. 564.84 is the human average, according to The Princeton Review.
Forwarded from Chat GPT (π
)
β**Daath AI Parser is an open-source application that uses OpenAI to parse visible text of HTML elements.**
GitHub
GitHub - serpapi/lego-ai-parser: Lego AI Parser is an open-source application that uses OpenAI to parse visible text of HTML elements.
Lego AI Parser is an open-source application that uses OpenAI to parse visible text of HTML elements. - serpapi/lego-ai-parser
Forwarded from Chat GPT (π
)
β)How Will AI Test Scientific Theories? Don't we need AI to be able to simulate and run experiments in order to get to AGI?
The prompt is simply a way to kick off a discussion. There is absolutely a lot of excitement surrounding the advances in OpenAI and DeepMind's systems. I think that what we're seeing with GPT will no doubt upend a lot of things that we humans do with regard to work, jobs, work product, even thinking to a certain extent.
But I do wonder if some of this hype (and even some of this fear-mongering going around) is disconnected from the other, significantly more challenging, directions that we need to push AI in order to get it where we need it to be for AGI (and, eventually, a technological singularity event).
AI can master language, verbal reasoning, and "fetch and build" types of creating (particularly, again, with regard to "writing things" using language), but can it, for example, conduct scientific tests in a simulated environment?
I think part of what will require creating an AGI will be programming an AI to be able to establish and simulate environments for testing. What I mean by that is: an AI needs to be able to understand everything that we know about our own understanding of physics, for example, AND be able to push the bounds of that understanding by running tests within its own software. The former I think is doable, we can do that just as we have done with language. But unlike with language, we want to rely on AI in these instances to make new discoveries, to add to our own understanding of physics. And then to use that to be creative and come up with new theories, it needs to be able to expand its programming in that way, to get better at testing. The hardware should ensure that tests will be done much, much faster, but how does the software get there?
I don't think we can have an AGI without that, can we?
In the meantime, the only Terminator scenario that AI can do right now is just produce misinformation. Until it can figure out to self-assemble (and it needs to push the bounds of our laws of physics, our understanding of chemistry, and in manufacturing processes in at least a couple of directions before it can do that), the worst AI can do is cause us to turn on each other. And that's an issue to be certain, but not of the type that I'm seeing coming from commentators and pundits.
But maybe I'm way off here. Would love to hear the sub's thoughts.
The prompt is simply a way to kick off a discussion. There is absolutely a lot of excitement surrounding the advances in OpenAI and DeepMind's systems. I think that what we're seeing with GPT will no doubt upend a lot of things that we humans do with regard to work, jobs, work product, even thinking to a certain extent.
But I do wonder if some of this hype (and even some of this fear-mongering going around) is disconnected from the other, significantly more challenging, directions that we need to push AI in order to get it where we need it to be for AGI (and, eventually, a technological singularity event).
AI can master language, verbal reasoning, and "fetch and build" types of creating (particularly, again, with regard to "writing things" using language), but can it, for example, conduct scientific tests in a simulated environment?
I think part of what will require creating an AGI will be programming an AI to be able to establish and simulate environments for testing. What I mean by that is: an AI needs to be able to understand everything that we know about our own understanding of physics, for example, AND be able to push the bounds of that understanding by running tests within its own software. The former I think is doable, we can do that just as we have done with language. But unlike with language, we want to rely on AI in these instances to make new discoveries, to add to our own understanding of physics. And then to use that to be creative and come up with new theories, it needs to be able to expand its programming in that way, to get better at testing. The hardware should ensure that tests will be done much, much faster, but how does the software get there?
I don't think we can have an AGI without that, can we?
In the meantime, the only Terminator scenario that AI can do right now is just produce misinformation. Until it can figure out to self-assemble (and it needs to push the bounds of our laws of physics, our understanding of chemistry, and in manufacturing processes in at least a couple of directions before it can do that), the worst AI can do is cause us to turn on each other. And that's an issue to be certain, but not of the type that I'm seeing coming from commentators and pundits.
But maybe I'm way off here. Would love to hear the sub's thoughts.
Reddit
From the singularity community on Reddit: Eliezer Yudkowsky has great news: "Parents conceiving today may have a fair chance ofβ¦
Explore this post and more from the singularity community
β€1
Forwarded from Chat GPT (π
)
Probably the worst database ever, but it can write poems!
I created a little wrapper I Python that allows you to convince ChatGPT that it's esentially a key-value store and use it for storage.
ChatGPT is an alternative to Redis that you can ask for poems in the middle of queries. Can your Redis write short poems?
The code is available here: https://github.com/styczynski/chatdb
(and Medium article describing approach in detail is accessible here)
Example of the query asking for poem in the middle of operations sequence
Imagine one of the state-of-the-art model being forced to emulate simple progra...
View original post
I created a little wrapper I Python that allows you to convince ChatGPT that it's esentially a key-value store and use it for storage.
ChatGPT is an alternative to Redis that you can ask for poems in the middle of queries. Can your Redis write short poems?
The code is available here: https://github.com/styczynski/chatdb
(and Medium article describing approach in detail is accessible here)
Example of the query asking for poem in the middle of operations sequence
Imagine one of the state-of-the-art model being forced to emulate simple progra...
View original post
Forwarded from Chat GPT (π
)
This media is not supported in your browser
VIEW IN TELEGRAM
simulating an operating system
π€―3