Az, there seems to be a misuse of terminology in this "new space". Defining or intimating this new software is artificial intelligence is misguided (not you, the 'collective' participants in the 'discussion' about WTF is "it". I believe this needs to be resolved ASAP.
These LLM are NOT human like intelligence or reasoning engines. This needs to be made clear.
An LLM is a statistical projection engine. It takes existing data elements and estimates the highest probably association the next term/phrase, in the context of the limit recall capacity defined by the training set.
This is nothing like a human. Sorry.
In fact, the concept of Artificial intelligence is so UNDEFINED, it's the convenient label to stick on anything AI related.
Az, this is noise.
What needs to happen is to deconstruct the elements of cognition (artifacts and problem solving skills) and execution. The results are measure precisely against the intended outcome or in the NEW creation which was previous UNASSOCIATED with this/these data points. In isolation, this is the "task" function of the LLM...and will be ONE of millions of solution sets based on experiential artifacts of billions of humans.
Problem: Comupter system (soup to nuts) do not have experiential artifacts to reference. As such, there is no circumstantial reference point for whatever actions followed to resolve the event. Yes, the path can be plotted and all the EXTRINSIC element can be 'captured' (vid, voice, location, time). What's missing is the creative element. Without a long diatribe here, let's stated the obvious.
AT PRESENT, LLM's are capable of highlevel logic inference and can make a variety of suggestions about "what's next". BUT, only in the context of a closed data set with a known set of constructs and qualified set of outcomes. In other words, if I want to solve the problem of why my wipers on my caddy are stuck, AND the key doesn't work AND the batt light is one. The probablity I am going to get THE answer will be inside a highly curated data set which has NO NOISE (like google search). Or? Yes, the gene fold data set is pure, OpenAI is not. That is where the state of the tech is today.
It produces good information, especially at this time when the hype cycle is about 40% reved up. Wait til Mar/Apr when the earth starts spin again. In that time, the only ideas which need attention are the technologies and solutions which REFINE the feed back loops, create new neurons insert in pre and post processing for clarity and front end interfaces which are EASIER than text or typing.
Today, this would be defined as an AGENT. Exactly. We are a colleciton of neural agents. Some drive the car (w/o thought) some cook perfect eggs (w/o thought), some resolve arguements (w creative visioning based on personal historical artifacts), etc, etc, etc.
This will be "STEP #2" on this path of technology innovation.
ALL the calls which state none sense like "Artificial Human intelligence that can mimic humans will be HERE in 2025"
No. Sam. Folks, that is Sams pitch deck. He is one of 5 people racing to the top Elon, Zuck, OpenAI, Goo, and Amaz. Who cares!!
The money, the near term money...I'm placing my bet on personal task centric agents where the results are measurable. (yes, thin on dets here, but from the 100,000 view)
We are a long way from human level artificial intelligence
I think one of the most profound developments will be AI undertaking knowledge work. So, not just a creator of knowledge, but a doer of work. Not just routine, drudge work but vast swathes of white collar work that provides wages for millions. This has very significant implications that are barely appreciated today by those who employ all us humans today.
Indeed. Already discovering some of that. Although, confession LLMs have for me made it easier to do higher value work than my drudge work because drudge work needs more precision....
Congrats on 500!
Az, there seems to be a misuse of terminology in this "new space". Defining or intimating this new software is artificial intelligence is misguided (not you, the 'collective' participants in the 'discussion' about WTF is "it". I believe this needs to be resolved ASAP.
These LLM are NOT human like intelligence or reasoning engines. This needs to be made clear.
An LLM is a statistical projection engine. It takes existing data elements and estimates the highest probably association the next term/phrase, in the context of the limit recall capacity defined by the training set.
This is nothing like a human. Sorry.
In fact, the concept of Artificial intelligence is so UNDEFINED, it's the convenient label to stick on anything AI related.
Az, this is noise.
What needs to happen is to deconstruct the elements of cognition (artifacts and problem solving skills) and execution. The results are measure precisely against the intended outcome or in the NEW creation which was previous UNASSOCIATED with this/these data points. In isolation, this is the "task" function of the LLM...and will be ONE of millions of solution sets based on experiential artifacts of billions of humans.
Problem: Comupter system (soup to nuts) do not have experiential artifacts to reference. As such, there is no circumstantial reference point for whatever actions followed to resolve the event. Yes, the path can be plotted and all the EXTRINSIC element can be 'captured' (vid, voice, location, time). What's missing is the creative element. Without a long diatribe here, let's stated the obvious.
AT PRESENT, LLM's are capable of highlevel logic inference and can make a variety of suggestions about "what's next". BUT, only in the context of a closed data set with a known set of constructs and qualified set of outcomes. In other words, if I want to solve the problem of why my wipers on my caddy are stuck, AND the key doesn't work AND the batt light is one. The probablity I am going to get THE answer will be inside a highly curated data set which has NO NOISE (like google search). Or? Yes, the gene fold data set is pure, OpenAI is not. That is where the state of the tech is today.
It produces good information, especially at this time when the hype cycle is about 40% reved up. Wait til Mar/Apr when the earth starts spin again. In that time, the only ideas which need attention are the technologies and solutions which REFINE the feed back loops, create new neurons insert in pre and post processing for clarity and front end interfaces which are EASIER than text or typing.
Today, this would be defined as an AGENT. Exactly. We are a colleciton of neural agents. Some drive the car (w/o thought) some cook perfect eggs (w/o thought), some resolve arguements (w creative visioning based on personal historical artifacts), etc, etc, etc.
This will be "STEP #2" on this path of technology innovation.
ALL the calls which state none sense like "Artificial Human intelligence that can mimic humans will be HERE in 2025"
No. Sam. Folks, that is Sams pitch deck. He is one of 5 people racing to the top Elon, Zuck, OpenAI, Goo, and Amaz. Who cares!!
The money, the near term money...I'm placing my bet on personal task centric agents where the results are measurable. (yes, thin on dets here, but from the 100,000 view)
We are a long way from human level artificial intelligence
What we have is
AUGEMENTED INTELLIGENCE.
and that's it for a while.
my thought.
I think one of the most profound developments will be AI undertaking knowledge work. So, not just a creator of knowledge, but a doer of work. Not just routine, drudge work but vast swathes of white collar work that provides wages for millions. This has very significant implications that are barely appreciated today by those who employ all us humans today.
Indeed. Already discovering some of that. Although, confession LLMs have for me made it easier to do higher value work than my drudge work because drudge work needs more precision....