Google CEO Sundar Pichai is on a PR blitz around artificial intelligence and the threat to the search giant. The tech is real. But there's also an ulterior motive.
Great piece, and congrats on the Politico article!
Many people in my life, personally and professionally, use ChatGPT to compose first drafts of communications of all sorts. Apologies for being so general, but I don't want to throw anyone under the bus.
I tried it out this year, and think it is VERY cool. I am a lifelong tech geek and normally adopt new tech into my life without hesitation. Except this time. I refuse to use it on ethical grounds. For around 15 years now, my first draft of almost all documents I create has been me talking into my phone. Before smart phones, it was me talking into a voice recorder.
I can see people who create mountains of content using it. I don't fault anyone in any industry or profession for using AI to create content (words to videos and everything else and in between).
However, I hope all do a thorough edit of those creations before they publish for wide consumption, and that a standard note of "AI was used to generate an initial copy of this content" is used for all such content.
The real problem with search apps is that they now are VERY BAD at search precisely because they are VERY GOOD at advertising. I go back very far in this professional space and in the early days of search there was a very robust search language that obeyed all Boolean operations, recognized delineated search strings and even field limitations, so that one could specify exactly the textual string, and permissible search results.
Now, and carried to an extreme with Google, all searches no matter how well crafted are first treated as a series of independent keywords and those keywords are first compared to relevant advertisements, such advertising presented first in the search results ahead of more relevant results, often pages ahead. And the is little left not obeyed of formerly robust search language.
I would much rather pay a monthly fee for a search service that was not on an advertising supported financial model.
My brother's wife called him the other day in a rage because she had to write a one page letter justifying a school field trip to get money from the district. My brother fired up ChatGPT with the prompt: "Write a 3 paragraph letter about why taking kids on a field trip to the air and space museum meets California state education standards". Two minutes later his wife had a brilliant letter and subsequently got her funding. There's a big future for companies that use AI to help people navigate government/corporate bureaucracies.
Not to nit-pick, but it is doubtful that ChatGPT could pass the Turing test, which requires that the average person be unable to tell, after five mintes conversation, whether they were talking to another person, or a machine.
Five minutes of conversation is a lot. While you might not notice the odd usage or misplaced metaphor if you weren't looking, most people would be able to make the correct evaluation with a little effort.
Since you asked: I am using ChatGPT 3.5 solely for exploratory purposes, to better understand its power and its potential applications. I have not yet experimented with the next generation ChatGPT 4.0. As a lawyer, my primary interest is seeing how it could be incorporated into a law firm’s practice. Can it help create first drafts of court motions and briefs, contracts, document summaries, correspondence, etc.? I just attended a symposium hosted by Silicon Flatirons in Boulder and was blown away by how blown away the presenters were by ChatGPT 4.0’s ability to draft a summary judgment motion/brief and a corporate merger agreement. They thought it provided an excellent “first draft,” subject to a seasoned lawyer’s further review, creative add-ons, and polishing. In the legal field, I felt the panel consensus was that ChatGPT would start to carry the lion’s share of performing some legal tasks, freeing up lawyers to focus on other things involving their creative processes, insights, and interpersonal connections: that it would augment the skills of, rather than replace, lawyers, and economize their time. Where I’m focused now, however, are the ethical implications of employing ChatGPT in one’s legal practice. These ethical implications are many and varied, and not always very obvious. Some involve an understanding of how ChatGPT works, and its baked in limitations and inherent biases.
An area where ChatAI is likely to have enormous implications is with regard to access to justice. Someday, sooner rather than later, non-lawyers will be able to obtain access to easy-to-understand ChatAI explanations of their legal rights (and others’ legal wrongs); be able to have ChatAI write cogent letters to companies and government agencies describing their complaints and pointing out potential legal violations; and, be able to represent themselves pro se armed with a plain English understanding of the pertinent legal claims and defenses, and to submit a legal brief in small claims or county court indistinguishable from one written by a graduate of one of our finer law schools.
Matt, I use Chat GPT to summarize issues for me either to expedite my writing or to understand issues/concepts that I have no familiarity with; for these uses, it is much better than a Google search. What doesn't work is finding academic or evidence-based research articles; ChatGPT rarely produces accurate results, sending me in a circular loop of finding that the sources aren't accurate, asking it to correct itself, getting more results, finding that those are inadequate and so on. I'm a non-fiction business book ghostwriter BTW.
I still think we ought to "nuke it from orbit, it's the only way to be sure" that AI doesn't become more evil than it already is.
It's been my feeling from the beginning that AI is too easily manipulated by those with less than honorable intentions. Start with the social biases inherent in the training tools used for the learning algorithms. If you're going to scrape the internet for training, you're going to get the cesspits along with the gems. Eventually the AI's will start talking about being human. Oh, wait...
And, look at all the different fields of endeavor that are being invaded by AI's like ChatGPT and the others. It's becoming ever more difficult to determine if a particular work is human-generated or comes from an AI. Thanks, but no thanks.
One aspect of the AI boom I don't see discussed much is the economics of the whole thing. ChatGPT-style queries are much more expensive to respond to. I have seen estimates of 10-times more compute power. We are in the typical tech land-rush stage now where user numbers are the gold and investors pile on the cash. How things shake out when the inevitable need to monetize things arrives will be interesting to see.
The technology is interesting, but google is an advertising platform and makes their money and exerts control through that, and it is much less amenable to disruption. Anti-monopoly should be focused there, on the grubby world of that, not on nebulous specifics of technology.
I had to quickly come up with a toast to the newlyweds at dinner. ChatGPT created something quite nice and utilitarian. While AI can compose a basic business letter and such, when it learns to hurl Shakespearean insults I will know the Rubicon has been crossed.
I have noodled around on chatGPT but haven’t used it in my day-to-day life. My 17 yo was struggling for an outline and starting his paper so I suggested he brainstorm it using chatGPT & it was helpful.
ChatGPT is now my most valuable tool as a programmer. I would give up StackOverflow and Google searches before I would give up ChatGPT. It’s basically having a very good developer who knows almost everything to work with.
And when it gives me a wrong answer - I get mad at it as I would at a fellow human.
I use it for work quite a bit. I use it to figure out basic issues in databases like oracle or Postgres. Or if I’m writing some small code and want to know how to do something it can be helpful.
It’s not always accurate at first so you have to keep asking questions to get the real answer.
One aspect of the AI land rush I don't see much about is its economics. ChatGPT queries are much more expensive to respond to. They take much more compute power than traditional style queries. I saw an estimate of a factor of 10 times more but I can't find the source now. We are in the land-rush stage now where user numbers are the gold and investors pile on the cash. That will not last forever and what happens when the suppliers face the need to make the business line profitable will be interesting to see.
Great piece, and congrats on the Politico article!
Many people in my life, personally and professionally, use ChatGPT to compose first drafts of communications of all sorts. Apologies for being so general, but I don't want to throw anyone under the bus.
I tried it out this year, and think it is VERY cool. I am a lifelong tech geek and normally adopt new tech into my life without hesitation. Except this time. I refuse to use it on ethical grounds. For around 15 years now, my first draft of almost all documents I create has been me talking into my phone. Before smart phones, it was me talking into a voice recorder.
I can see people who create mountains of content using it. I don't fault anyone in any industry or profession for using AI to create content (words to videos and everything else and in between).
However, I hope all do a thorough edit of those creations before they publish for wide consumption, and that a standard note of "AI was used to generate an initial copy of this content" is used for all such content.
That would make me content.
The real problem with search apps is that they now are VERY BAD at search precisely because they are VERY GOOD at advertising. I go back very far in this professional space and in the early days of search there was a very robust search language that obeyed all Boolean operations, recognized delineated search strings and even field limitations, so that one could specify exactly the textual string, and permissible search results.
Now, and carried to an extreme with Google, all searches no matter how well crafted are first treated as a series of independent keywords and those keywords are first compared to relevant advertisements, such advertising presented first in the search results ahead of more relevant results, often pages ahead. And the is little left not obeyed of formerly robust search language.
I would much rather pay a monthly fee for a search service that was not on an advertising supported financial model.
My brother's wife called him the other day in a rage because she had to write a one page letter justifying a school field trip to get money from the district. My brother fired up ChatGPT with the prompt: "Write a 3 paragraph letter about why taking kids on a field trip to the air and space museum meets California state education standards". Two minutes later his wife had a brilliant letter and subsequently got her funding. There's a big future for companies that use AI to help people navigate government/corporate bureaucracies.
Great column, as usual. Lots to think about.
Not to nit-pick, but it is doubtful that ChatGPT could pass the Turing test, which requires that the average person be unable to tell, after five mintes conversation, whether they were talking to another person, or a machine.
Five minutes of conversation is a lot. While you might not notice the odd usage or misplaced metaphor if you weren't looking, most people would be able to make the correct evaluation with a little effort.
Since you asked: I am using ChatGPT 3.5 solely for exploratory purposes, to better understand its power and its potential applications. I have not yet experimented with the next generation ChatGPT 4.0. As a lawyer, my primary interest is seeing how it could be incorporated into a law firm’s practice. Can it help create first drafts of court motions and briefs, contracts, document summaries, correspondence, etc.? I just attended a symposium hosted by Silicon Flatirons in Boulder and was blown away by how blown away the presenters were by ChatGPT 4.0’s ability to draft a summary judgment motion/brief and a corporate merger agreement. They thought it provided an excellent “first draft,” subject to a seasoned lawyer’s further review, creative add-ons, and polishing. In the legal field, I felt the panel consensus was that ChatGPT would start to carry the lion’s share of performing some legal tasks, freeing up lawyers to focus on other things involving their creative processes, insights, and interpersonal connections: that it would augment the skills of, rather than replace, lawyers, and economize their time. Where I’m focused now, however, are the ethical implications of employing ChatGPT in one’s legal practice. These ethical implications are many and varied, and not always very obvious. Some involve an understanding of how ChatGPT works, and its baked in limitations and inherent biases.
An area where ChatAI is likely to have enormous implications is with regard to access to justice. Someday, sooner rather than later, non-lawyers will be able to obtain access to easy-to-understand ChatAI explanations of their legal rights (and others’ legal wrongs); be able to have ChatAI write cogent letters to companies and government agencies describing their complaints and pointing out potential legal violations; and, be able to represent themselves pro se armed with a plain English understanding of the pertinent legal claims and defenses, and to submit a legal brief in small claims or county court indistinguishable from one written by a graduate of one of our finer law schools.
Matt, I use Chat GPT to summarize issues for me either to expedite my writing or to understand issues/concepts that I have no familiarity with; for these uses, it is much better than a Google search. What doesn't work is finding academic or evidence-based research articles; ChatGPT rarely produces accurate results, sending me in a circular loop of finding that the sources aren't accurate, asking it to correct itself, getting more results, finding that those are inadequate and so on. I'm a non-fiction business book ghostwriter BTW.
I still think we ought to "nuke it from orbit, it's the only way to be sure" that AI doesn't become more evil than it already is.
It's been my feeling from the beginning that AI is too easily manipulated by those with less than honorable intentions. Start with the social biases inherent in the training tools used for the learning algorithms. If you're going to scrape the internet for training, you're going to get the cesspits along with the gems. Eventually the AI's will start talking about being human. Oh, wait...
And, look at all the different fields of endeavor that are being invaded by AI's like ChatGPT and the others. It's becoming ever more difficult to determine if a particular work is human-generated or comes from an AI. Thanks, but no thanks.
One aspect of the AI boom I don't see discussed much is the economics of the whole thing. ChatGPT-style queries are much more expensive to respond to. I have seen estimates of 10-times more compute power. We are in the typical tech land-rush stage now where user numbers are the gold and investors pile on the cash. How things shake out when the inevitable need to monetize things arrives will be interesting to see.
Saving to read this later. I suspect this is when Stoller breaks my heart. TBC
The technology is interesting, but google is an advertising platform and makes their money and exerts control through that, and it is much less amenable to disruption. Anti-monopoly should be focused there, on the grubby world of that, not on nebulous specifics of technology.
I had to quickly come up with a toast to the newlyweds at dinner. ChatGPT created something quite nice and utilitarian. While AI can compose a basic business letter and such, when it learns to hurl Shakespearean insults I will know the Rubicon has been crossed.
I have noodled around on chatGPT but haven’t used it in my day-to-day life. My 17 yo was struggling for an outline and starting his paper so I suggested he brainstorm it using chatGPT & it was helpful.
ChatGPT is now my most valuable tool as a programmer. I would give up StackOverflow and Google searches before I would give up ChatGPT. It’s basically having a very good developer who knows almost everything to work with.
And when it gives me a wrong answer - I get mad at it as I would at a fellow human.
It totally changes how I develop software.
I use it for work quite a bit. I use it to figure out basic issues in databases like oracle or Postgres. Or if I’m writing some small code and want to know how to do something it can be helpful.
It’s not always accurate at first so you have to keep asking questions to get the real answer.
One aspect of the AI land rush I don't see much about is its economics. ChatGPT queries are much more expensive to respond to. They take much more compute power than traditional style queries. I saw an estimate of a factor of 10 times more but I can't find the source now. We are in the land-rush stage now where user numbers are the gold and investors pile on the cash. That will not last forever and what happens when the suppliers face the need to make the business line profitable will be interesting to see.
Holy crap, will it tell me when to Go?