Nov 12, 2022·edited Nov 12, 2022Liked by Alberto Romero
>> I wouldn’t be surprised if GPT-4 could pass it (the Turing test) amply.
If this is true -- with the provisos of a) doing so as well as humans with an IQ of at least 100
and b) at a cost level of not more than $10/hour -- this will be one of great technological revolutions in history. The applications are virtually infinite. *If it is true* society needs to start thinking about the consequences immediately. Because they will be enormous.
The marketplace has barely absorbed and leveraged the capabilities of existing SOTA models such as GPT-3. If GPT-4 turns out to be a quantum leap in capability, that gap will only widen. VC's, start-ups and incumbents will be scrambling to implement these untapped abilities. It will be interesting to see how much GPT-4 simplifies the ability to use it, accelerating the innovation that can be quickly rolled out.
I'm rather new to this and I'm struggling to understand how someone can know on Aug 20 what it'll be capable of, when they hadn't even started training it yet at that point?
It might pass Turing for short-attention-span theater topics, but if it doesn't have STM and LTM, a grade-school kid will out it.
The long context window with pre-fix padding salient extracted facts is a neat parlor trick, but it's still just 'deception'.
Still, one has to wonder whether the model has started to re-read the corpus with understanding, and based on that understanding, is able to 'fine-tune' it's knowledge in an auto-self-organization.
'Eventually, some AI systems will have both great technical ability and the social ability needed to understand human speech and wishes. If given charge of energy, materials, and assemblers, such a system might aptly be called a "genie machine." What you ask for, it will produce, Arabian legend and universal common sense suggest that we take the dangers of such engines of creation very seriously indeed.' -- Eric Drexler
If a novice AI reader is not entirely sure what this speculation is about the following article seems, as best I (definitely not an expert) can tell, to provide a pretty good introduction to the subject of GPT. It was helpful to me at least.
The speculation that most interests me is, what's going to happen when this kind of AI can write expert level articles that can compete successfully with those written by humans? My impression so far is that the expert writers community seems to be somewhat in denial in regards to this.
If the topic comes up at all, the threat to writers seems to usually be dismissed with "AI is not that advanced yet, it will take some time etc" Ok, fair enough, sounds reasonable. For now. But it's going to happen at some point, yes? No?
>> I wouldn’t be surprised if GPT-4 could pass it (the Turing test) amply.
If this is true -- with the provisos of a) doing so as well as humans with an IQ of at least 100
and b) at a cost level of not more than $10/hour -- this will be one of great technological revolutions in history. The applications are virtually infinite. *If it is true* society needs to start thinking about the consequences immediately. Because they will be enormous.
The marketplace has barely absorbed and leveraged the capabilities of existing SOTA models such as GPT-3. If GPT-4 turns out to be a quantum leap in capability, that gap will only widen. VC's, start-ups and incumbents will be scrambling to implement these untapped abilities. It will be interesting to see how much GPT-4 simplifies the ability to use it, accelerating the innovation that can be quickly rolled out.
Fascinating. This must be what it feels like to isolated primitives when an aircraft is sighted in the distance.
But what if this post is written by GPT-4? Or this comment?
I'm rather new to this and I'm struggling to understand how someone can know on Aug 20 what it'll be capable of, when they hadn't even started training it yet at that point?
The future is going to be wild!
It might pass Turing for short-attention-span theater topics, but if it doesn't have STM and LTM, a grade-school kid will out it.
The long context window with pre-fix padding salient extracted facts is a neat parlor trick, but it's still just 'deception'.
Still, one has to wonder whether the model has started to re-read the corpus with understanding, and based on that understanding, is able to 'fine-tune' it's knowledge in an auto-self-organization.
'Eventually, some AI systems will have both great technical ability and the social ability needed to understand human speech and wishes. If given charge of energy, materials, and assemblers, such a system might aptly be called a "genie machine." What you ask for, it will produce, Arabian legend and universal common sense suggest that we take the dangers of such engines of creation very seriously indeed.' -- Eric Drexler
as a programmer my job is in danger.
chatgpt will soon kill search engines
The applications are endless, I hope an AI like this is implemented in a game which makes the npcs more life-like.
If a novice AI reader is not entirely sure what this speculation is about the following article seems, as best I (definitely not an expert) can tell, to provide a pretty good introduction to the subject of GPT. It was helpful to me at least.
https://www.techtarget.com/searchenterpriseai/definition/GPT-3
The speculation that most interests me is, what's going to happen when this kind of AI can write expert level articles that can compete successfully with those written by humans? My impression so far is that the expert writers community seems to be somewhat in denial in regards to this.
If the topic comes up at all, the threat to writers seems to usually be dismissed with "AI is not that advanced yet, it will take some time etc" Ok, fair enough, sounds reasonable. For now. But it's going to happen at some point, yes? No?
A friend suggests that GPT-4 -- again, on the assumption of Turing Test equivalence -- might be a great tool to teach foreign languages.