It\u2019s\u00a0similar to\u00a0where we are\u00a0with AI today.\u00a0Like the resort, today\u2019s\u00a0users\u00a0of\u00a0AI assume\u00a0it\u2019s\u00a0cheaper to use AI\u00a0without restraint than to have humans do\u00a0what\u00a0the AI can do.\u00a0There\u2019s\u00a0a\u00a0notion that AI is more efficient\u00a0than humans for any and all\u00a0things. That there\u00a0is no\u00a0need\u00a0to\u00a0worry about\u00a0or even talk about usage cost. But that begs the question of whether that will always be the case.\u00a0<\/p>\n
Few vendors\u00a0and\u00a0users\u00a0are even thinking about\u00a0what will happen if,\u00a0and when,\u00a0the usage cost goes up\u00a0and goes up significantly.\u00a0Much less\u00a0talking\u00a0about it. Not at ILTACON. Not any place.<\/p>\n
Except perhaps\u00a0NetDocuments.<\/p>\n
The\u00a0<\/strong>L<\/strong>egal\u00a0<\/strong>C<\/strong>ontext\u00a0<\/strong>E<\/strong>ngineering\u00a0<\/strong>B<\/strong>enchmark\u00a0<\/strong>R<\/strong>eport\u00a0<\/strong><\/p>\n Back in May,\u00a0NetDocuments released a\u00a0Legal\u00a0Context\u00a0Graph<\/a>\u00a0and, more recently, a\u00a0Legal\u00a0Context\u00a0Engineering\u00a0Benchmark\u00a0Report<\/a>\u00a0about what the Graph showed.\u00a0Both the\u00a0Graph and\u00a0Report\u00a0were\u00a0designed to measure how better\u00a0context\u00a0and content\u00a0can make AI\u00a0use\u00a0more efficient and results\u00a0more\u00a0accurate\u00a0at less cost. The\u00a0Graph and\u00a0Report\u00a0purported to\u00a0show that the cost of obtaining a correct AI answer\u00a0can\u00a0be substantially reduced by adding content and context.<\/p>\n It\u2019s perhaps easy to dismiss\u00a0what NetDocuments is doing\u00a0here\u00a0since it is in the business of document management.\u00a0And it\u2019s in its own interest to tout\u00a0being able to use\u00a0increased context\u00a0to\u00a0increase\u00a0accuracy at a\u00a0lower\u00a0cost.\u00a0<\/p>\n But\u00a0that would be a mistake. What NetDocuments has done\u00a0is important\u00a0because\u00a0virtually\u00a0no one has\u00a0gone to the trouble of looking hard\u00a0at the relationship between\u00a0AI\u00a0cost and accuracy.\u00a0And\u00a0being able to determine\u00a0that ratio is\u00a0about to become\u00a0critical.<\/p>\n Today\u2019s Usage<\/strong>\u00a0<\/strong>Pattern<\/strong><\/p>\n Dan\u00a0Hauck, NetDocuments Chief Product Officer, describes\u00a0the\u00a0current\u00a0prompting process\u00a0this way:\u00a0a\u00a0user\u00a0types in a\u00a0prompt\u00a0in\u00a0an\u00a0AI platform\u00a0hoping\u00a0to get\u00a0the\u00a0answer\u00a0they are looking for. If they don\u2019t, they\u00a0simply\u00a0continue to prompt the tool\u00a0over and over\u00a0with additional information\u00a0until they get what they believe to be an accurate\u00a0answer.\u00a0They can do\u00a0that\u00a0because the cost of repeated prompting is typically not much of a factor,\u00a0and there\u2019s no way to undertake a cost-benefit analysis in any event.<\/p>\n But What About\u00a0Tomorrow\u2019s Usage, And\u00a0Its\u00a0Cost<\/strong><\/p>\n But\u00a0that\u2019s about to change. Here\u2019s why:\u00a0AI\u00a0vendors are investing millions, if not billions, of dollars in developing AI models and platforms. They\u00a0generally\u00a0charge\u00a0for these platforms on a subscription basis while losing\u00a0boatloads\u00a0of money. But\u00a0the\u00a0investors\u00a0in these vendors\u00a0are not going to let them get away with that\u00a0forever. At some point,\u00a0they will\u00a0demand that the vendors\u00a0show a profit.\u00a0A significant profit.<\/p>\n Indeed, one vendor,\u00a0Legora,\u00a0is\u00a0already\u00a0talking<\/a>\u00a0about\u00a0going to a consumption-based model.\u00a0The more\u00a0tokens\u00a0you use,\u00a0the more\u00a0the\u00a0cost\u00a0will be. Under that\u00a0kind of\u00a0model, the ability to\u00a0evaluate\u00a0the benefit of the AI usage for a task against the cost\u00a0could become\u00a0important.<\/p>\n And other\u00a0vendors\u00a0across the AI world\u00a0will no doubt follow. Why?\u00a0Because quite simply, they can.\u00a0Customers\u00a0are relying more and more\u00a0on AI\u00a0platforms. They are\u00a0changing\u00a0their\u00a0workflows and processes.\u00a0Some are even changing their business models.<\/p>\n All that makes it\u00a0very\u00a0hard\u00a0for them\u00a0to leave the AI\u00a0Garden of Eden, even if the usage cost goes\u00a0up\u00a0dramatically. It is a perfect recipe for changing\u00a0the pricing\u00a0model, charging\u00a0more for\u00a0tokens, and raising the\u00a0overall\u00a0price.\u00a0<\/p>\n Think\u00a0Amazon. In the beginning, Amazon charged a remarkably low cost\u00a0for its\u00a0eBooks.\u00a0In doing so, it cultivated a huge customer base over the years\u00a0while also losing money.\u00a0But once that base was all in\u00a0with eBooks,\u00a0once we all\u00a0bought\u00a0our\u00a0Kindles,\u00a0Amazon\u00a0began raising the price of those\u00a0eBooks\u00a0significantly. In doing so,\u00a0it\u00a0became extremely profitable. AI vendors will soon\u00a0adopt a similar strategy. As Josh\u00a0Baxter,\u00a0NetDocuments\u00a0CEO,\u00a0succinctly\u00a0told me\u00a0while we\u00a0were\u00a0at ILTACON:\u00a0\u201cThe reality is it\u2019s coming.\u201d<\/p>\n It\u2019s a reality that will require application of cost-benefit\u00a0analysis.<\/p>\n The Future Importance of the Cost<\/strong>\u2013<\/strong>Benefit Ratio<\/strong><\/p>\n That\u2019s why what NetDocuments is doing is important, even if perhaps it\u2019s before its time. It\u2019s providing a framework for\u00a0injecting\u00a0an accurate\u00a0cost-benefit analysis into the AI use matrix.\u00a0Here\u2019s what that could mean: when\u00a0increasing\u00a0the quality\u00a0and\u00a0accuracy\u00a0of an AI\u00a0output\u00a0by\u00a01%\u00a0costs\u00a010-20% more, the user may decide it\u2019s not worth it.\u00a0<\/p>\n I\u2019ve\u00a0experienced\u00a0this with the\u00a0insurance industry. Over time, carriers figured out that they could pay\u00a0notoriously\u00a0low rates to insurance defense lawyers, knowing that\u00a0incremental\u00a0increases\u00a0in quality\u00a0from\u00a0using\u00a0higher-priced lawyers\u00a0was\u00a0simply\u00a0not worth the increased cost.\u00a0My resort figured that out as well.<\/p>\n Yes, sometimes the increased\u00a0cost for increased accuracy may indeed be worth it. Sometimes it won\u2019t. But without some way to determine that\u00a0relationship, it would just be guesswork. And\u00a0the\u00a0temptation to be right\u00a0at whatever cost will be the default.<\/p>\n Baxter\u00a0puts it this way:\u00a0\u201cAs AI shifts to consumption pricing, the industry is swinging between two extremes: spending without limits and cutting without strategy.\u201d\u00a0With no way to figure it out.<\/p>\n It\u2019s Not Always Rock and Roll<\/strong><\/p>\n All of which is to say\u00a0that\u00a0NetDocuments is\u00a0taking\u00a0a step in the right direction. It\u2019s\u00a0tossing a\u00a0dry towel\u00a0on\u00a0all the drunken AI gushing that\u2019s going on.<\/p>\n And\u00a0I\u2019m not surprised that\u00a0it\u2019s\u00a0NetDocuments\u00a0that\u2019s\u00a0doing this.\u00a0I keep coming back to\u00a0what\u00a0Baxter\u00a0told me<\/a>\u00a0one time:\u201cWe\u2019re not a rock band.\u201d\u00a0That\u2019s what\u00a0distinguishes\u00a0the company in\u00a0a\u00a0sea of AI over the top hype.<\/p>\n And sometimes it\u2019s nice to listen to some quiet jazz instead of\u00a0Yungblud. Whoever\u00a0that is. It\u2019s the name ChatGPT gave me when I asked who the most over-the-top rock star is today.\u00a0<\/p>\n At least his name isn\u2019t Harvey.<\/p>\n Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes\u00a0TechLaw Crossroads<\/a>, a blog devoted to the examination of the tension between technology, the law, and the practice of law.<\/em><\/strong><\/p>\n The post The Cost of AI Usage:\u00a0Time To Throw In\u00a0The\u00a0Unrestrained\u00a0Towel\u00a0<\/a> appeared first on Above the Law<\/a>.<\/p>\n I used to vacation with my family and friends at a swanky resort in a foreign country. The resort had several grand pools and overlooked the Pacific. It also had a very strict towel policy. To get a towel, you had to sign in with an attendant who recorded your room number, name, and the time that you took the towel. At the end of the day, you had to return the towel to the attendant and have him write down the time it was brought in. If you didn\u2019t turn it in at the end of the day, the attendant reported that,\u00a0and\u00a0you were charged\u00a0a significant fee.\u00a0<\/p>\n Compound\u00a0that by several pools and God knows\u00a0how many towels,\u00a0and you\u00a0need\u00a0a lot of human labor\u00a0just to take care of the towels.\u00a0The cost\u00a0to pay all those attendants at all those pools\u00a0had to be a lot.\u00a0I\u00a0naively wondered at the time\u00a0whether it\u00a0would be\u00a0cheaper for the resort\u00a0to just absorb the cost of missing towels or\u00a0somehow\u00a0use technology to replace\u00a0part of what all those attendants did.\u00a0<\/p>\n Then it dawned on me.\u00a0At the\u00a0meager\u00a0wage levels of workers in that\u00a0country, it was\u00a0much\u00a0cheaper\u00a0to\u00a0pay\u00a0humans than to eat the cost of lost towels or buy the technology to replace them.<\/p>\n But as labor costs went up, all that changed. Now you grab your towel\u00a0from a bin,\u00a0and if you remember to turn it in, you just drop it in another\u00a0bin on the way out. If you forget, no\u00a0big deal.<\/p>\n T<\/strong>oday\u2019s Cost of AI Use<\/strong><\/p>\n It\u2019s\u00a0similar to\u00a0where we are\u00a0with AI today.\u00a0Like the resort, today\u2019s\u00a0users\u00a0of\u00a0AI assume\u00a0it\u2019s\u00a0cheaper to use AI\u00a0without restraint than to have humans do\u00a0what\u00a0the AI can do.\u00a0There\u2019s\u00a0a\u00a0notion that AI is more efficient\u00a0than humans for any and all\u00a0things. That there\u00a0is no\u00a0need\u00a0to\u00a0worry about\u00a0or even talk about usage cost. But that begs the question of whether that will always be the case.\u00a0<\/p>\n Few vendors\u00a0and\u00a0users\u00a0are even thinking about\u00a0what will happen if,\u00a0and when,\u00a0the usage cost goes up\u00a0and goes up significantly.\u00a0Much less\u00a0talking\u00a0about it. Not at ILTACON. Not any place.<\/p>\n Except perhaps\u00a0NetDocuments.<\/p>\n The\u00a0<\/strong>L<\/strong>egal\u00a0<\/strong>C<\/strong>ontext\u00a0<\/strong>E<\/strong>ngineering\u00a0<\/strong>B<\/strong>enchmark\u00a0<\/strong>R<\/strong>eport\u00a0<\/strong><\/p>\n Back in May,\u00a0NetDocuments released a\u00a0Legal\u00a0Context\u00a0Graph<\/a>\u00a0and, more recently, a\u00a0Legal\u00a0Context\u00a0Engineering\u00a0Benchmark\u00a0Report<\/a>\u00a0about what the Graph showed.\u00a0Both the\u00a0Graph and\u00a0Report\u00a0were\u00a0designed to measure how better\u00a0context\u00a0and content\u00a0can make AI\u00a0use\u00a0more efficient and results\u00a0more\u00a0accurate\u00a0at less cost. The\u00a0Graph and\u00a0Report\u00a0purported to\u00a0show that the cost of obtaining a correct AI answer\u00a0can\u00a0be substantially reduced by adding content and context.<\/p>\n It\u2019s perhaps easy to dismiss\u00a0what NetDocuments is doing\u00a0here\u00a0since it is in the business of document management.\u00a0And it\u2019s in its own interest to tout\u00a0being able to use\u00a0increased context\u00a0to\u00a0increase\u00a0accuracy at a\u00a0lower\u00a0cost.\u00a0<\/p>\n But\u00a0that would be a mistake. What NetDocuments has done\u00a0is important\u00a0because\u00a0virtually\u00a0no one has\u00a0gone to the trouble of looking hard\u00a0at the relationship between\u00a0AI\u00a0cost and accuracy.\u00a0And\u00a0being able to determine\u00a0that ratio is\u00a0about to become\u00a0critical.<\/p>\n Today\u2019s Usage<\/strong>Pattern<\/strong><\/p>\n Dan\u00a0Hauck, NetDocuments Chief Product Officer, describes\u00a0the\u00a0current\u00a0prompting process\u00a0this way:\u00a0a\u00a0user\u00a0types in a\u00a0prompt\u00a0in\u00a0an\u00a0AI platform\u00a0hoping\u00a0to get\u00a0the\u00a0answer\u00a0they are looking for. If they don\u2019t, they\u00a0simply\u00a0continue to prompt the tool\u00a0over and over\u00a0with additional information\u00a0until they get what they believe to be an accurate\u00a0answer.\u00a0They can do\u00a0that\u00a0because the cost of repeated prompting is typically not much of a factor,\u00a0and there\u2019s no way to undertake a cost-benefit analysis in any event.<\/p>\n But What About\u00a0Tomorrow\u2019s Usage, And\u00a0Its\u00a0Cost<\/strong><\/p>\n But\u00a0that\u2019s about to change. Here\u2019s why:\u00a0AI\u00a0vendors are investing millions, if not billions, of dollars in developing AI models and platforms. They\u00a0generally\u00a0charge\u00a0for these platforms on a subscription basis while losing\u00a0boatloads\u00a0of money. But\u00a0the\u00a0investors\u00a0in these vendors\u00a0are not going to let them get away with that\u00a0forever. At some point,\u00a0they will\u00a0demand that the vendors\u00a0show a profit.\u00a0A significant profit.<\/p>\n Indeed, one vendor,\u00a0Legora,\u00a0is\u00a0already\u00a0talking<\/a>\u00a0about\u00a0going to a consumption-based model.\u00a0The more\u00a0tokens\u00a0you use,\u00a0the more\u00a0the\u00a0cost\u00a0will be. Under that\u00a0kind of\u00a0model, the ability to\u00a0evaluate\u00a0the benefit of the AI usage for a task against the cost\u00a0could become\u00a0important.<\/p>\n And other\u00a0vendors\u00a0across the AI world\u00a0will no doubt follow. Why?\u00a0Because quite simply, they can.\u00a0Customers\u00a0are relying more and more\u00a0on AI\u00a0platforms. They are\u00a0changing\u00a0their\u00a0workflows and processes.\u00a0Some are even changing their business models.<\/p>\n All that makes it\u00a0very\u00a0hard\u00a0for them\u00a0to leave the AI\u00a0Garden of Eden, even if the usage cost goes\u00a0up\u00a0dramatically. It is a perfect recipe for changing\u00a0the pricing\u00a0model, charging\u00a0more for\u00a0tokens, and raising the\u00a0overall\u00a0price.\u00a0<\/p>\n Think\u00a0Amazon. In the beginning, Amazon charged a remarkably low cost\u00a0for its\u00a0eBooks.\u00a0In doing so, it cultivated a huge customer base over the years\u00a0while also losing money.\u00a0But once that base was all in\u00a0with eBooks,\u00a0once we all\u00a0bought\u00a0our\u00a0Kindles,\u00a0Amazon\u00a0began raising the price of those\u00a0eBooks\u00a0significantly. In doing so,\u00a0it\u00a0became extremely profitable. AI vendors will soon\u00a0adopt a similar strategy. As Josh\u00a0Baxter,\u00a0NetDocuments\u00a0CEO,\u00a0succinctly\u00a0told me\u00a0while we\u00a0were\u00a0at ILTACON:\u00a0\u201cThe reality is it\u2019s coming.\u201d<\/p>\n It\u2019s a reality that will require application of cost-benefit\u00a0analysis.<\/p>\n The Future Importance of the Cost<\/strong>\u2013<\/strong>Benefit Ratio<\/strong><\/p>\n That\u2019s why what NetDocuments is doing is important, even if perhaps it\u2019s before its time. It\u2019s providing a framework for\u00a0injecting\u00a0an accurate\u00a0cost-benefit analysis into the AI use matrix.\u00a0Here\u2019s what that could mean: when\u00a0increasing\u00a0the quality\u00a0and\u00a0accuracy\u00a0of an AI\u00a0output\u00a0by\u00a01%\u00a0costs\u00a010-20% more, the user may decide it\u2019s not worth it.\u00a0<\/p>\n I\u2019ve\u00a0experienced\u00a0this with the\u00a0insurance industry. Over time, carriers figured out that they could pay\u00a0notoriously\u00a0low rates to insurance defense lawyers, knowing that\u00a0incremental\u00a0increases\u00a0in quality\u00a0from\u00a0using\u00a0higher-priced lawyers\u00a0was\u00a0simply\u00a0not worth the increased cost.\u00a0My resort figured that out as well.<\/p>\n Yes, sometimes the increased\u00a0cost for increased accuracy may indeed be worth it. Sometimes it won\u2019t. But without some way to determine that\u00a0relationship, it would just be guesswork. And\u00a0the\u00a0temptation to be right\u00a0at whatever cost will be the default.<\/p>\n Baxter\u00a0puts it this way:\u00a0\u201cAs AI shifts to consumption pricing, the industry is swinging between two extremes: spending without limits and cutting without strategy.\u201d\u00a0With no way to figure it out.<\/p>\n It\u2019s Not Always Rock and Roll<\/strong><\/p>\n All of which is to say\u00a0that\u00a0NetDocuments is\u00a0taking\u00a0a step in the right direction. It\u2019s\u00a0tossing a\u00a0dry towel\u00a0on\u00a0all the drunken AI gushing that\u2019s going on.<\/p>\n And\u00a0I\u2019m not surprised that\u00a0it\u2019s\u00a0NetDocuments\u00a0that\u2019s\u00a0doing this.\u00a0I keep coming back to\u00a0what\u00a0Baxter\u00a0told me<\/a>\u00a0one time:\u201cWe\u2019re not a rock band.\u201d\u00a0That\u2019s what\u00a0distinguishes\u00a0the company in\u00a0a\u00a0sea of AI over the top hype.<\/p>\n And sometimes it\u2019s nice to listen to some quiet jazz instead of\u00a0Yungblud. Whoever\u00a0that is. It\u2019s the name ChatGPT gave me when I asked who the most over-the-top rock star is today.\u00a0<\/p>\n At least his name isn\u2019t Harvey.<\/p>\n
\n![\"\"](\"https:\/\/i0.wp.com\/abovethelaw.com\/wp-content\/uploads\/sites\/4\/2024\/07\/Netdocuments-logo-300x207.jpg?resize=300%2C207&ssl=1\" "\\"\\""){kind=logo}
<\/figure>\n
\n