YouTube transcript
Video ID: ME_WXT9oboE
Language: en
Captions: auto-generated

[0:00] All of a sudden like the contracts and
[0:02] and all the memos all sounds like I
[0:06] written the content without real good
[0:08] editing and people didn't put enough
[0:11] effort in it. I always tell my users
[0:13] best way to eliminate AI is really write
[0:19] yourself. The best humanizer is
[0:21] ourselves. People underestimate how
[0:24] complex our brain works. Like most of
[0:27] the professional writing
[0:29] especially
[0:31] technical it's also very creative right
[0:34] now like I simp it doesn't have that
[0:37] level intelligence it it can do the
[0:39] repetitive work well it can do technical
[0:42] work well it can do like pure creative
[0:45] things well like but but it but if you
[0:48] mix all that up it it really suck. Hey
[0:51] everybody, I'm Devin Kums. I'm a CPA,
[0:53] MBA, and I work at Google Cloud, where I
[0:55] lead multi-billion dollar deal
[0:57] negotiations as a financial expert. But
[1:00] for fun, I also do this podcast,
[1:02] Ambition Aligned, where I try to find
[1:03] purpose-driven professionals making real
[1:06] impact with what they're doing in their
[1:08] lives. Please watch, please subscribe,
[1:10] and I think we're going to get a ton out
[1:11] of today's episode. Tell me what's been
[1:13] going on with you, my friend. Well,
[1:15] what's new? Yeah, I mean it's a lot
[1:17] going on like
[1:19] between law school and you know and my
[1:22] day job and then all the startups it's
[1:25] it's you know it's a
[1:27] lot.
[1:29] So I I think when you run the startup
[1:32] really you have to take care of
[1:33] everything from scratch like like I code
[1:37] I code myself. I do the testing myself
[1:40] and also do the marketing reach out to
[1:42] people and do the demos uh and and also
[1:45] I helping a lot of companies you know
[1:48] doing kind of automations right right so
[1:52] you know quite of quite a lot of like
[1:54] kind of workflows uh basically I build I
[1:56] need to build from scratch people don't
[1:58] have those automations in house yet so I
[2:01] do a lot of those with quite a few uh
[2:05] tech companies in the Bay Area and plus
[2:07] I'm bio science companies um like really
[2:12] in the United States. So I think right
[2:15] now three of my customers are going the
[2:19] for the IPO. Uh so the world volume
[2:23] really ps up and on top of that you have
[2:26] like up to like law schools my exams are
[2:28] coming up. So that's that part isn't
[2:31] easy either. So well that saying goes
[2:34] right like if you want to get something
[2:36] done give it to a busy
[2:38] person that you epitomize that. So tell
[2:41] us tell me more about I want to hear
[2:42] more about the law school and how you're
[2:44] balancing all this. But first what's
[2:46] going on with the startup? What are you
[2:48] specifically working on? Three IPOs in a
[2:50] pretty dry IPO market. So that's pretty
[2:52] impressive. What's going on there? Well,
[2:54] you know I you know I have two shops
[2:56] right like Chipmunk Robotics which is my
[2:58] consulting shop like who help people out
[3:01] in terms of implementing automation that
[3:04] with people establish the sock controls
[3:07] uh
[3:08] and making sure their recognitions are
[3:11] correct like being all over the place
[3:12] with chipm robotics. Uh and the second
[3:15] startup I have is really uh joint
[3:17] venture with a number of engineers uh
[3:20] which is uh AI startup for professional
[3:23] drafting inkwise AI. Uh so we on board
[3:27] with close to like 1,000 uh businesses
[3:32] so far and most of it are small
[3:34] business.
[3:35] Um most of those users are professional
[3:38] attorneys uh and technical folks as well
[3:43] as finance investment folks are in
[3:46] different areas in consulting uh or in
[3:49] psychological those type of areas too.
[3:52] uh but that's that get me busy in terms
[3:56] of uh you know people how to leverage
[3:59] the platform to to you know to drive
[4:02] professional doctors on those platforms
[4:05] right so that's almost like completely
[4:07] different things from what I've been
[4:10] doing
[4:12] robotics yeah tell me more about inkwise
[4:15] so what what are you doing at inkwise
[4:16] now and what is walk me through the
[4:18] workflow like what exactly it seems like
[4:21] you're pretty excited about it And so
[4:22] I'm curious if you want to share more.
[4:24] Yeah, absolutely. I mean, inquire is
[4:26] essentially you can you can view inquire
[4:28] as being an automation as well. It's
[4:31] automation for professional writing by
[4:33] leveraging uh
[4:35] JI at this point. Everybody uses um GI
[4:40] now like everybody uses either TGBT or
[4:44] anthropic and and probably you and I we
[4:47] both use TGBT and that's to do you know
[4:50] certain amount of writing whether it's
[4:53] professional writing or not like that's
[4:54] a different story uh but I think I use
[4:57] TBT to help me to do circus um and but
[5:02] at the end of day often times uh you
[5:05] know I copy what GPT or ASOP told me and
[5:09] pasted in a word dog and then I started
[5:11] organizing my sauce and building things
[5:13] together and because AI generated
[5:17] content are first of all like you can't
[5:21] blend like fast right and then secondly
[5:24] yeah hallucinations yeah like and even
[5:26] even without hallucination you know the
[5:29] way I write
[5:31] and you can tell like two like when
[5:35] people send few stuff read. You can tell
[5:38] it's the written stuff, right? Oh, right
[5:41] away. Right. The um dash, the um dash
[5:43] and then the rocket ships, the emojis,
[5:45] the amount of emojis going on right now.
[5:47] It's ridiculous. And then what else is
[5:49] that giveaway? Yeah. Like this. But
[5:52] yeah, there's a lot of like tropes,
[5:54] right? Yeah. This is the part this is
[5:56] one of the reason I built inquires is
[5:59] well first of all I saw like like a lot
[6:01] of professionals like us like we start
[6:03] using
[6:05] robic in more of a silly way like where
[6:08] people just put generate content
[6:10] directly in the professional document
[6:12] impact memos in contracts and all of a
[6:16] sudden like the contracts and and all
[6:19] the memos all sounds
[6:22] like I written in the content without uh
[6:26] real good editing. Uh and people didn't
[6:29] put enough effort in it. uh and partly
[6:31] is because I mean like the typical way
[6:35] to handle it uh like with based you know
[6:38] technology like Microsoft word uh you
[6:40] know some of version come with copilot
[6:43] and Google Google doc comes come up you
[6:46] know comes with gym power uh AI
[6:50] assistance as well but even like you al
[6:53] have to copy what I generated for you
[6:57] and paste in there and then you can try
[6:59] to do editing and it interrupts your
[7:04] thought process. Think about in old ways
[7:07] when we
[7:08] write we open a blank word document and
[7:12] we open a blank Google doc document and
[7:14] we just start writing there right like
[7:15] our flow is is not interrupted
[7:19] constantly by something like
[7:21] AI that that's what I'm trying to build
[7:24] using in that
[7:27] is you know I we have to acknowledge is
[7:30] is here and everyone wants to use it but
[7:33] it saves us time you know us
[7:35] successfully for especially for those
[7:38] mechanical part of the right uh but in
[7:41] is built to give you the ability to
[7:44] write together with AI so more seamless
[7:49] uh like pull AI like I don't use a
[7:52] chatbot experience I don't have a
[7:53] chatbot uh uh like well people start
[7:57] writing they can write as is or writing
[8:00] in you know word document and whenever
[8:04] they need That's where can use the my my
[8:08] like prompt window is more like a inline
[8:11] prompter can button to bring the
[8:14] prompter and this is where they can ask
[8:16] AI to write based on the reference
[8:19] document they populed. So that's not
[8:22] different
[8:24] from a handling of the references
[8:27] document too. Uh and that's where we're
[8:29] getting into a little bit technical.
[8:33] So when when like people right now they
[8:36] can upload reference document name to
[8:38] check is for for those solutions to
[8:41] write based on those references
[8:44] right professional document think about
[8:46] like company filing for IPO is drafting
[8:49] a entire T like S1 or or T or 10K 10
[8:54] it's a long document right your
[8:56] reference
[8:58] support you can have hundreds of
[9:00] references document or you know tens
[9:03] support document in there. It could be
[9:05] Excel, could be a word, could be PDF all
[9:08] those kind of different types of
[9:09] document but some of the document can
[9:12] like three of the document has support
[9:14] this one paragraph the other two
[9:16] supports another one and there's and
[9:19] there might be 10 reference document
[9:21] supports another section right so it's
[9:24] not it's not like all the document you
[9:26] uploaded to AI can be used all at the
[9:29] same time when you do the right
[9:32] So and and isn't smart enough like after
[9:36] upload let's say tens of document into
[9:38] uh into chatpt like it's not smart
[9:42] enough to tell like to to basically
[9:44] distinguish what are writing for this
[9:46] argument for this paragraph this
[9:48] paragraph need which part of the
[9:51] references uh you just upload right like
[9:54] like AI isn't just like us yeah it's far
[9:57] away from being us in in reality so
[10:01] That's why you know I created inquire uh
[10:04] we we use a totally like more uh more
[10:08] sophisticated sophisticated rag
[10:10] technology through augmented generation
[10:13] to basically allow users to pick from
[10:17] their references documents like say they
[10:20] can pick those two reference document
[10:22] for this paragraph and they can pick
[10:23] another document for another paragraph.
[10:26] So that human element of it like I'm
[10:29] trying to
[10:31] essentially transform the writing
[10:34] experience from the existing natural
[10:37] writing habit we in ways to integrate
[10:42] with AI as writing like I'm still
[10:44] leveraging AI but I don't change our
[10:48] existing for for us writing a papers you
[10:51] know you teach teach at the college
[10:53] right so you give paper assignments to
[10:55] your students When the students write
[10:58] those papers, they have all those
[11:00] supporting documents or all those like
[11:02] journals they reference in the research,
[11:04] right? They they basically read those
[11:07] papers and I write the conclusion or
[11:11] write things based on those papers they
[11:14] just same thing like like forest writing
[11:18] uh that shouldn't change. Uh people
[11:22] shouldn't just be lazy or I'm I'm
[11:24] dumping everything in there and just
[11:26] hoping AI can pick and just pick
[11:29] everything for me like it's first of all
[11:32] is not smart enough to do that like
[11:34] secondly that takes away the human
[11:37] thinking process from from the red so
[11:40] and that's where I don't don't like to
[11:43] see that like for for being a
[11:45] professional uh you know for being a CPA
[11:48] and and also conflict like can be a
[11:50] future lawyer as well. Uh, you know, I
[11:53] definitely don't want to see the
[11:55] professionals are are really on AI and
[11:58] stop
[12:00] stop knowing how to write themselves.
[12:04] Yeah, it's so interesting because I've
[12:06] played in the AI space and I want to ask
[12:08] more about Inkwise and what you're using
[12:09] there because it's just so interesting
[12:10] to me right now and it sounds like
[12:12] you're making great tractions with it
[12:13] and I'm curious too on how you're
[12:14] pricing it and all that. There's a
[12:16] million things I want to ask you, but
[12:18] what I've seen is because I've played
[12:20] with AI probably way too much and for a
[12:23] bunch of different use cases because I
[12:24] just love tinkering with technology and
[12:26] along with a lot of the fine-tuned
[12:27] models around it. Like TA bot is one of
[12:29] them that I had on recently that's
[12:30] pretty solid.
[12:32] U it really isn't that good. Like I
[12:35] guess that's my big takeaway. It's
[12:37] really good at a paragraph. About a
[12:39] paragraph of context you can get. I like
[12:41] can you help me polish this paragraph?
[12:43] Can you tell me if I missed anything
[12:45] important or is there anything you can
[12:46] It's really good with brainstorming too.
[12:48] Like how did this go? What am I missing?
[12:50] Can you give me some other ideas? But
[12:52] then if as soon as I get to a page or
[12:54] even multi-page article, I could give a
[12:57] very explicit guidance and I'm talking
[12:59] about the best models on chat GBT or um
[13:01] Claude and be like here's the things
[13:03] like what one of my workflows I like to
[13:05] do is I used to like to go on walks and
[13:06] make voice notes for myself and I'll get
[13:09] very clear like here's how I want an
[13:10] article. I want to site this, this,
[13:11] this, these books, and normally then I
[13:14] would take that and go write. And I'm
[13:15] like, hey, can chat GPT go do the
[13:17] writing for me or or any of these tools.
[13:19] They can't. It just it just hands down
[13:22] can't. It'll be way fake. It'll miss a
[13:24] bunch of the sources. It'll me miss a
[13:26] bunch of the details of the story. It
[13:27] makes the writing experience actually
[13:28] more painful. Um, at best it could be
[13:31] just like outlining and like it makes
[13:32] you not want to write at all. You just
[13:34] want to just be like, "This wasn't what
[13:36] I was going for at all." And it uses all
[13:38] of those tropes. So, um I was curious
[13:40] like how does how do you enhance that
[13:42] workflow and do you use audio at all or
[13:44] have you seen anything that uses audio?
[13:47] Well,
[13:49] well, I before I built like a quest
[13:54] honestly myself I I didn't want to build
[13:56] it because I know how many products on
[14:00] the market doing those kind of AI
[14:02] writing.
[14:04] uh because writing is supposed to be
[14:06] what AI is best at because they're
[14:11] they're genetic like there are large
[14:13] language models right this is what's
[14:15] supposed to be what they're really good
[14:17] at uh and and there are tons of
[14:19] solutions out there like I kind like at
[14:22] least 30 of for technical accounting you
[14:25] know how how much like technical
[14:27] accounting levels I I wrote so
[14:30] um not like just like you said I had
[14:33] choose like none of those works well. Uh
[14:37] well, from the start, most of the start
[14:41] with a chatbot experience. You kind of
[14:44] write stuff like you could kind of write
[14:46] like professional serious content with
[14:49] chatbot experience, right? It's it's
[14:53] like it's it's really
[14:55] counter what writing is. Like writing is
[15:00] um iterative
[15:03] and process which you know writers have
[15:06] to go in their mind like you know round
[15:09] and round of editing in their mind
[15:11] before they even start
[15:13] writing writers like good writer need to
[15:15] know what they want to achieve you know
[15:17] before they even put keyboard uh ready
[15:21] for so but but the chatbot is taking
[15:24] that away from people chatbot
[15:28] is p with AI and it gave views a one
[15:34] shot uh result of that tag and that
[15:40] itself is basically the the very large
[15:45] roadblock for you to to anyway like to
[15:49] achieve a very good writing uh result
[15:52] out of that experience. So even for the
[15:55] ones that is not a tech experience, I
[15:59] feel people like the the engineer
[16:02] building those solutions are relying on
[16:04] AI too much.
[16:07] Um I think I think AI could be could be
[16:12] like basically good at either good at
[16:16] very very creative stuff. uh for like
[16:19] for example you said AI can help us
[16:21] brainstorm a lot of things with those
[16:22] things that's actually that's the
[16:25] creative part and AI is also very good
[16:28] at uh very very technical
[16:33] uh thing themselves like like for
[16:35] example like AI I think the best AI use
[16:38] cases today and most value added
[16:40] services AI is doing today is help
[16:44] engineers engineer themselves to code
[16:47] Right. And and and the is actually doing
[16:50] a very good you know like showing us a
[16:53] very good result in there. Nowadays you
[16:55] can go to some AI platform and
[16:57] generating a front page like front app
[17:01] like like like web applications like you
[17:05] know in minutes just using it as help.
[17:08] So I think on both sides AI is doing
[17:10] very well. I think when it comes down to
[17:13] writing um people like underestimate how
[17:17] complex our brain works like most of the
[17:21] professional writing
[17:23] especially
[17:24] technical it's also very creative you
[17:28] know it's a very very subtle way you
[17:31] make it all your technical writing work
[17:35] to make it very persuasive you can make
[17:38] it like works a very different way like
[17:41] the changes like our human brain are
[17:45] making is is very subtle right now like
[17:49] AI simply doesn't have that level of
[17:52] intelligence like close to us right it
[17:55] can do the repetitive work well it can
[17:58] do technical work well it can do like
[18:01] pure creative things well like but but
[18:04] it but if you mix all that
[18:06] up it it really suck it really
[18:10] So that I think that's that's one of the
[18:12] things like at least for a lot of
[18:15] technical writers out there. Uh I I
[18:18] heard those kind of rumors a lot from
[18:21] people like oh you just give it another
[18:24] five years AI is going to replace the
[18:26] attorneys. I is going to replace most of
[18:29] the contents
[18:32] uh maybe to some degree to some degree
[18:36] because a lot of repetitive work that
[18:39] they can be replaced by AI but I think
[18:42] at the
[18:44] moment it's proven
[18:48] current AI algorithms by simply feeding
[18:52] a large data set won't get us far
[18:57] enough at to the level of the human
[19:01] intelligence. Doesn't matter how many
[19:04] you
[19:05] know there are popular 10 you know
[19:09] primary models out
[19:11] there that that level like you can feed
[19:15] all kinds of content to AI still doesn't
[19:19] work like a human brain work. Um, yeah,
[19:23] I'm meeting a lot of people. I'm meeting
[19:24] a lot of companies who are realizing
[19:26] this and are really investing in having
[19:28] someone like I've heard this term
[19:30] service as a software a lot lately of
[19:33] like the idea is you need someone to
[19:36] manage the a professional a really
[19:37] highly paid professional to manage the
[19:40] model and the output and maybe they have
[19:42] a special fine-tuned model that supports
[19:44] them in that output. But you need a
[19:46] human. You you need a human for that
[19:48] final leg. And it almost reminds me of
[19:50] like how we all have the you know when
[19:53] you call us a call center now you all
[19:54] have the automated center and everyone
[19:56] was so big like CCAI call center AI was
[19:59] such a big move 10 years ago but I think
[20:02] we can all admit that it made the world
[20:03] worse like no one no one wants to talk
[20:06] to that freaking robot on the other side
[20:08] right like everyone's instantly pressing
[20:10] the be like give me a representative
[20:12] give me a representative and I could see
[20:15] if we don't manage this correctly or
[20:17] businesses don't manage it correctly
[20:19] this could become another solution like
[20:20] that. But it sounds like you're trying
[20:21] to solve that with um with Inguise. So
[20:23] what what are you doing differently?
[20:25] Well, well I I'm not saying like my
[20:28] technology or my build is more is better
[20:31] than other people's. I'm just taking
[20:33] more of a human approach.
[20:36] Like I I think like I often tell people
[20:38] like inquisit is probably the the most
[20:42] anti-
[20:43] AI rising platform called like people
[20:47] going there
[20:49] expecting oh the I have a chatbot and
[20:52] basically I ask the chat I ask AI in the
[20:55] chat like hey help me write a tactical
[20:57] event or help me write a contract about
[20:59] certain things. That's not what inquest
[21:02] is. in give you a blank page and give
[21:06] you a platform for you to write and then
[21:09] you can you can pull AI whenever you
[21:12] want to to help you finish a paragraph
[21:14] help you finish synthesis people needs
[21:17] to know how to to use the platform like
[21:20] inquest it's it's essentially no
[21:23] different from the old ways of writing
[21:25] using Microsoft word or using you know
[21:28] Google doc so it doesn't change any uh
[21:31] it basically fast track Your workflows
[21:33] allow you to upload the template and
[21:35] allow you to quickly roll forward,
[21:37] but it requires people know how to write
[21:41] the things they want to write and it
[21:43] requires people to still spend the time
[21:46] to find what references they need to
[21:49] support the writing and they requires
[21:51] people to know okay which references is
[21:54] used for to support which paragraph. So
[21:56] it it basically have all that human
[21:58] element in there. Um but this is where
[22:03] uh they can also pull in AI's help to
[22:06] fill the mechanical gaps. This is where
[22:09] like AI can come in to help or quickly
[22:12] quickly write a argument based on the
[22:14] three references you know people select.
[22:18] So that's I think that's the beauty of
[22:20] inquisite. That's where um it doesn't
[22:23] change anyone's writing habit and also
[22:27] doesn't allow AI to dominate the
[22:30] narrative. So uh
[22:33] it it really can let people to dominate
[22:36] their
[22:39] writing. No, that's that's so
[22:41] fascinating. I'm trying to think like a
[22:42] practical application and this is how I
[22:44] would do it too. is I'm going to write a
[22:45] memo, give some of my company
[22:47] background, make sure I know what
[22:49] sources I'm citing, and then start going
[22:52] into like what's the objective of this
[22:53] memo? How am I going to accomplish this
[22:55] task? What arguments are for and are
[22:57] against it? And you're saying there's
[22:59] more of a companion, like a maybe like a
[23:00] co-pilot, but I'm in the middle of my
[23:03] paragraph. I'm like, "Hey, I think I
[23:04] gave you enough here. Can you write
[23:06] finish this up for me?" And it can do
[23:07] that. Is that kind of how you're using
[23:09] it or how do you use it in your
[23:10] workflow? And how are you finding that
[23:12] your workflow is improving from it? Uh,
[23:14] do you want me to show you how it works?
[23:16] Yeah. Could you show it? Let's go. Yeah,
[23:18] if you don't mind. Yeah. Yeah. Yeah. Uh,
[23:21] can let me let me get it in. I show you
[23:23] how it works. So,
[23:25] uh, so I don't just stop. Uh, don't
[23:28] always just talk. So,
[23:32] uh, we're like both of us are 606
[23:34] experts. So, maybe let's talk about six.
[23:38] Let's do it, man. Second. Let's I
[23:40] appreciate. And you're saying a thousand
[23:41] people are using this. Is this what your
[23:43] IPOs are doing or are you doing that
[23:44] through the chipmunk? Uh so there are
[23:47] about like a thousand users uh small
[23:49] business most uh are using inquest
[23:52] already. Um are they paid or beta? Oh
[23:55] there are most of them are already so Oh
[23:58] wow that's awesome. Yeah I mean
[24:01] um and this whole thing you know when I
[24:04] build the the automations it it's also
[24:07] not expensive right. So I are like very
[24:11] like moderate pricing. Uh I would say
[24:15] yeah your your rates are always steals.
[24:16] Yeah. Uh here here is 606 contract
[24:20] review manual right like and what I did
[24:23] is okay I have a sales audit form which
[24:25] is my contract in here. I have some SSP
[24:28] documentation in there and I have some
[24:30] my personal 606 guidance accumulated
[24:34] over the years in there right how me
[24:36] support to write this memo. So uh there
[24:39] are there are things like uh and you
[24:42] mentioned about predictive like uh
[24:44] autocomplete right and we all know how
[24:46] autocompletes work right let me let me
[24:50] show you how autocompletes works in like
[24:53] in today's world like using AI right now
[24:56] and you see this is allocation so I I
[24:59] listed the SSPs and I listed the
[25:02] contract price and and AI is automat
[25:05] like automatic autocompletes will
[25:08] consider all of the references I have.
[25:10] It will also consider the existing
[25:13] content like what I'm trying to achieve
[25:16] here and it also consider the whole art
[25:19] like whole positions of the article the
[25:21] structure of the article and basically
[25:24] there is a whole um I don't know I
[25:26] talked to a lot of folks um people
[25:28] mentioned about AI agent there's a whole
[25:30] AI agent working behind the scene to
[25:32] make this happen.
[25:35] Yeah. and and and this is cool. No, I'm
[25:38] just reading it right now. This is
[25:39] amazing. This is so much better than
[25:42] what I've seen chat bots would ever do.
[25:44] Yeah. And this is where I can imagine is
[25:46] there a verbal way to do this? Like is
[25:47] there a way that I could talk into this
[25:48] and then you could structure it because
[25:50] it's or I guess you could at least copy
[25:53] paste into Inkwise at the very least.
[25:55] But there's something really cool here.
[25:56] This is something I'm always looking
[25:57] for. It's like, hey, here's my idea and
[26:00] now I don't want to write all the BS
[26:01] that gets it to the end. Like the main
[26:03] idea and arguments there. Can something
[26:04] just finish it for me? But if I was to
[26:06] put this in chat GBT or Claude or
[26:08] Gemini, it would give me something
[26:09] completely different and incorrect. But
[26:11] this is following the right flow and
[26:13] it's because you basing are the agents
[26:16] going against the documents that you've
[26:18] uploaded to compare against or what's
[26:19] going on there. Exactly. It's it's first
[26:21] references all the references I I
[26:24] uploaded myself and it's also look at
[26:27] okay what's what what are the preceding
[26:30] immediately proceeding things I wrote so
[26:33] far then it also consider okay this what
[26:37] this section is about and what is this
[26:40] whole article is about and what is this
[26:42] section's relationship with the rest of
[26:44] the sections in this article so it
[26:47] consider multiple things before it
[26:49] generated this content text. So like
[26:52] here I said total contract price. So we
[26:55] already have a total contract price. The
[26:58] allocation for cloud subscription is
[27:00] here. So then what's remaining like in
[27:03] this contract what's remaining is just a
[27:05] professional services piece right and
[27:08] then it's essentially basically know
[27:11] what's remaining what I want to say is
[27:14] the remaining of the contract price
[27:16] allocated to pay professional services
[27:19] and so that way you know we like
[27:24] basically by doing all those AI agent
[27:26] behind the scenes we basically
[27:29] engineered AI to be able to predict what
[27:32] I want to write next. And and the beauty
[27:36] about this is the more you write in this
[27:40] article in this
[27:41] document, the more accurate that will be
[27:45] for the next
[27:47] uncomplete. And and of course, if you
[27:49] don't like this version, right, you can
[27:51] always try a different version. Uh you
[27:54] just press the arrow key and try a
[27:55] different version uh for AI to show it
[27:58] to you. And both scanning both are
[28:00] solid. All of them have been great ways
[28:02] to end of it. Exactly. So and and and
[28:05] this is just one one feature that
[28:07] autocomplete feature is just the one
[28:09] feature helping people to quickly draft
[28:11] the document. Uh the main inline
[28:14] prompter you can you can see the whole
[28:16] thing is just the whole thing is just
[28:19] the
[28:19] editor. Uh you you can start with
[28:22] templates here. You don't have to start
[28:25] with templates. Um and what you can do
[28:28] like lost people is if you just want to
[28:31] roll forward certain levels you simply
[28:34] just select this and this this is where
[28:37] you see where I where I right click this
[28:39] is the inline
[28:41] prompt this is where you can start
[28:44] asking AI to write certain paragraphs
[28:47] for for you like here let's say this
[28:50] pallet pallet term description is old
[28:53] right I just want AI to use this
[28:56] contract and this is where it's
[28:58] different from TBD. I allow people to up
[29:01] to upload the references but auto select
[29:03] which reference to use for which
[29:05] paragraph and say
[29:07] rewrite the selected
[29:11] text based on the
[29:15] out in the reference.
[29:18] Oh, this is awesome. and and you can
[29:21] this is where AI is only going to
[29:24] retrieve that information from the
[29:26] references isolate. It doesn't go to go
[29:30] look at the other two references and it
[29:33] also knows okay user selected text is
[29:36] the pallet term so they know it needs to
[29:39] write the pallet term and also it's
[29:41] going to show me where did they got the
[29:44] information from the pallet term from
[29:46] the original contract. So, oh, so it
[29:49] actually cites it. That's amazing. Yeah,
[29:51] cuz you know, you don't want to
[29:52] hallucinate and you know, to make so and
[29:57] this is where the pallet terms and and
[30:00] if I if I don't like it, I actually ask
[30:03] can I ask it and try it again? Uh, and
[30:05] it like can try many times if I want.
[30:07] Uh, and then if I'm okay with this, I
[30:10] can just quickly edit it here and do
[30:12] some quick editing.
[30:14] um just delete those old ones and and
[30:17] here we go. Right? That's a quick quick
[30:19] roll forward. Uh people can do um but
[30:24] but besides a roll forward, I still
[30:26] prefer has a you know has a first user
[30:29] on this platform. I still prefer people
[30:31] to to write on their own and use AI
[30:35] whenever whenever we need to. Right. So
[30:38] you know and how's the math? So can we
[30:40] go back to that section you did? I'm
[30:42] curious how the good the math is because
[30:43] I always find that AI is really rough
[30:46] with the math. So it's you are
[30:48] definitely correct. Absolutely correct.
[30:50] AI is not good at math. So like for
[30:54] example here uh remember this numbers
[30:57] for five if it try to ask AI to
[31:02] calculate this number. Let's say let's
[31:05] say whether I can calculate this number
[31:09] correctly. You see this is wrong.
[31:12] Yeah. Yeah. Because very interesting. So
[31:14] you can't rely on the numbers. So it's
[31:16] still a struggle. Yeah. Don't don't
[31:18] don't this is where don't rely on don't
[31:22] completely real AI to do things
[31:24] especially like in professional writing
[31:26] like it's it's not it's not intelligent
[31:29] enough to replace us. At least not yet.
[31:32] Well, I found it really I found it
[31:34] really funny. I've seen like some AI
[31:35] enhanced financial planning tools. I'm
[31:38] like it's giving you wrong numbers. It's
[31:40] the worst. It's not meant to do this.
[31:43] Yeah. Exactly. Exactly.
[31:46] And I think people uh people really need
[31:49] to be careful especially when they
[31:51] obtaining a lot of contracts and and and
[31:56] ultimately ultimately it's it's a person
[32:00] delivering those document are
[32:02] responsible for what they wrote, right?
[32:05] So it's not AI is not going to be
[32:06] responsible for. So you know I I think I
[32:10] think this is where um like we need we
[32:13] can use AI to do a lot of lot of help
[32:16] like for example if I I have this
[32:19] argument I have this conclusion here and
[32:22] my conclusion for the pilot because
[32:24] their trial period right my conclusion
[32:27] is the contract will not be recognized
[32:29] until the pallet period concludes. So,
[32:32] so I I have many pre-built AI tools
[32:37] basically I can I can ask AI to write
[32:40] opposing
[32:42] argument
[32:44] immediately for like from the from the
[32:48] paragraph I selected and this is this is
[32:50] the where AI is leveraging the 660 and
[32:53] trying to say there is a contract
[32:56] um so
[32:59] that's cool this is awesome Yeah, this
[33:01] is so good. It was I was I was just
[33:03] talking to someone else about how
[33:05] important it is to like one of the
[33:08] biggest risks about AI especially for
[33:09] memo writing software is it does it it
[33:12] can constrain the brainstorming. You
[33:14] want to be able to use it to brainstorm.
[33:15] And if it gives you an answer, you might
[33:17] take it for face value, especially if
[33:18] you're not an expert and logical. And
[33:20] I'm always looking for like, hey, first
[33:23] humanize it. Like, can you write this
[33:24] like you're not a robot? And then
[33:26] second, could like what what's the gaps
[33:29] in the logic here? Or what what am I
[33:30] missing? And this is huge. I'm curious.
[33:32] And how accurate is it? I'm kind of want
[33:34] to Well, the earlier the conclusion is
[33:37] my conclusion. So, you know, my
[33:39] conclusion was correct one.
[33:41] I when I ask to write the counter
[33:45] argument it's meant to say well
[33:47] firstly I focus on this uh not solely on
[33:52] termination clauses it tried to refute
[33:55] my argument there that's cool and it it
[33:58] physically and and secondly the existent
[34:01] refor can be assessed beyond the
[34:03] termination clause you see like it's
[34:05] really trying to build the right
[34:07] argument along the way trying to beat me
[34:10] uh when I in there. Um uh
[34:13] and
[34:15] so yeah, so like here like at the end of
[34:19] the day like they don't all the man
[34:21] preclude RA recognition during health
[34:24] care.
[34:26] I think all he's saying is correct. It's
[34:29] just it's just basically trying to say
[34:31] companies make Kai recognized revenue in
[34:34] accordance even before head period
[34:37] concludes. And so you can you can see
[34:39] how
[34:41] how much that's interesting. Yeah, you
[34:43] can see how much AI is trying its best
[34:45] to beat my to beat my conclusion here.
[34:49] Um and and and it's actually very very
[34:52] helpful for a lot of attorneys uh trying
[34:54] to draft uh uh like a legal argument um
[34:58] and they when they whenever they put the
[35:01] argument in there they're they can use
[35:03] AI to see okay can AI be and like do
[35:08] they have a new angle uh and miss you
[35:12] know they they can actually use this and
[35:14] and then you mentioned the humanizer
[35:15] like you know I I definitely have a
[35:17] humanizer as myself like trying to
[35:20] trying to eliminate the AI elements of
[35:23] the writing from the existing AI
[35:26] generated content. Uh but but I always
[35:30] tell my users like the best way to
[35:33] eliminate
[35:36] AI style is really write yourself
[35:41] right don't rely on AI writing like like
[35:45] after just use AI generation for summary
[35:48] and just re like rewrite on top of that
[35:50] just using it and fill the mechanical
[35:52] gaps. The best the best humanizer is
[35:55] human like ourselves. How funny is that?
[35:59] It's like so many people are trying to
[36:00] get AI to write like humans. Like just
[36:02] hey, how about you finish the writing?
[36:04] How about you do the last polish? Don't
[36:06] get too lazy, right? Yeah. Exactly.
[36:08] Exactly. So, well, what so what are how
[36:12] are you selling this or how what's the
[36:13] plan? How are you marketing all of this?
[36:15] Um are how are you balancing this with
[36:17] all the other stuff that you're working
[36:18] on? Is this your priority right now or
[36:20] what's going on? Um absolutely not a
[36:22] priority. Um so for the most part
[36:25] inquire is is getting popularities among
[36:28] professionals but it's really I I didn't
[36:31] do any marketing myself. Uh it's really
[36:34] you know mouse to mouse right now at the
[36:37] moment. So you know what the content use
[36:40] this like oh this is really great and
[36:43] they tell their friends and same same
[36:45] thing with the attorneys. Um I started
[36:48] this I started this originally for
[36:50] technical account like us. Um but but
[36:53] right now I actually fact that Laur
[36:56] attorney is on the platform data
[36:58] content. Um oh interesting. Um so we the
[37:02] words is really uh go fast right now
[37:05] like uh you know I this is not just in
[37:08] the United States you know I have users
[37:11] from across the globe right now using
[37:13] this uh because you know large language
[37:15] models like doesn't really matter which
[37:18] language
[37:20] uh is important right um so yeah I
[37:25] really have people from globe using this
[37:27] uh to
[37:28] correct this is awesome Boy, I know I
[37:31] know I've seen it before, but I now want
[37:33] to I appreciate you doing the live demo
[37:34] for me. I want to I want to play with it
[37:36] now. I want to play with it a lot more
[37:37] and I'm sure some of the viewers will
[37:39] too. So, I'll make sure to put a link in
[37:40] for everybody and I know you get free
[37:42] demos out, right? So, the um yeah, this
[37:44] is awesome. Thank you for sharing. It's
[37:46] really exciting. So, are you what are
[37:48] you doing on are you still building this
[37:50] out more or what is what does your
[37:51] dayto-day look like now or what what are
[37:53] you excited about next for all of this
[37:55] and all of your projects? you know by in
[37:57] in the next uh actually let me I will
[38:00] show you this uh I'm still building some
[38:03] of the features uh especially the team
[38:05] feature so
[38:08] uh this is my long long-term dream like
[38:12] where
[38:13] um where like multiple people kind of
[38:17] work on a very long technical document
[38:19] for example take I think is a great
[38:22] example and you know we we've Being a uh
[38:27] technical content ourselves, we know how
[38:29] long how lenzy the 10K is and how far it
[38:32] is to draft the annual report for any
[38:35] public companies out there, right? Like
[38:38] like imagine imagine multiple people use
[38:42] can basically use AI use references to
[38:45] write different sections of uh technical
[38:48] document like 10K and 10Q at the same
[38:52] time. Um so it's
[38:55] really
[38:56] substantially increase the efficiencies
[39:00] of those
[39:02] document certain menus certain menus or
[39:05] certain uh like even even you want to
[39:08] draft a long book or like you can
[39:10] basically put all your references in
[39:13] there and different section of the book
[39:15] all at the same time. It's
[39:17] really this is really a game changer.
[39:21] um when one person use AI to write you
[39:24] know it probably can reduce the time by
[39:27] like 80% easy uh when multiple people
[39:31] can use AI to draft certain things at
[39:34] the same time and that's really that's
[39:37] really where the return uh really come
[39:39] in like like historically for uh for S1
[39:44] like for for companies going to going
[39:47] IPO right the S1 drafting It's a very
[39:51] very painful process and it it can take
[39:55] weeks. Uh and and really this like once
[39:58] I have that's putting it lightly. It
[40:00] could take months and you could be
[40:01] sleeping in a printing house for weeks.
[40:03] Yeah. I've been there. Yeah. Exactly. So
[40:05] and you know I have multiple team
[40:08] members can use uh the rack the AI uh to
[40:11] draft uh technical sections very
[40:14] accurately and then this is this is the
[40:17] way it elevate the writing to the next
[40:20] level and it's also like like the
[40:24] writing. So this is like to me like this
[40:28] is not a new concept uh but it's a it's
[40:31] a new it's really a new technology in
[40:34] the AI space where you know where people
[40:38] can use shared drag to write uh a
[40:42] document a very technical lens document
[40:46] um by multiple people together and and
[40:48] this is this is almost
[40:51] like a a burst of Google doc
[40:55] all over again. Uh just just in the age
[40:59] of AI, right? So that's Well, it's so
[41:02] fascinating that Yeah, it's so
[41:04] fascinating that you can compete so much
[41:05] better than a Google doc with this like
[41:07] that your functionality just seems so
[41:09] much cleaner and more useful than in
[41:11] Google Doc. I can't do any of this right
[41:13] now. I just get the chatbot Gemini
[41:15] inclusion. Yeah. Not because Google
[41:17] doesn't have a technology, it doesn't
[41:19] have because more of uh like like here
[41:25] um um like when I when I go in
[41:31] um when I go into the platform when I do
[41:35] the upload like for example uh when I
[41:37] write this 606 uh memo I upload my
[41:40] references each of my references each my
[41:44] document it says it's a word document
[41:46] here. But in reality, each of the
[41:49] document I upload, I actually convert
[41:52] each document into a mini actor
[41:54] databases
[41:56] people like like for example whenever I
[41:59] call this prompter here, I select this
[42:02] contract and in fact selecting the
[42:06] database vector database converted from
[42:09] this reference document.
[42:12] So and and for a large company like
[42:15] Google think about probably for Google
[42:18] or for you know Microsoft is as they
[42:22] have Google drive like Microsoft has one
[42:25] drive like if you want to support
[42:28] technical writing using rag solution
[42:30] like this they have to convert how many
[42:34] documents and the like they basically
[42:37] just think about just just you alone how
[42:40] many documents you have in your Google
[42:41] drive right If you want to convert each
[42:44] of the document to a database that
[42:47] Google
[42:48] manages, how much volume Google has to
[42:51] deal with? Yeah, it's impossible. Yes,
[42:53] it's it's fascinating. It's it's a
[42:55] classic big big fish problem. It's just
[42:57] too much. Like this is where why we need
[43:00] competitors in the space like yourself
[43:02] because you have that advantage of being
[43:03] lean and being able to not overprocess.
[43:06] Man, I appreciate that. We're running
[43:08] close on time. We got to talk more about
[43:10] all the other stuff you're working on.
[43:13] Um what's what's next to mind for you
[43:16] other than this? Like
[43:19] well I I definitely want to push uh a
[43:22] number of things. Uh like I feel a
[43:24] little guilty like this year I didn't do
[43:27] many publications. I used to do uh in
[43:31] automation. I used to publish articles
[43:33] telling people hey how to automate uh
[43:36] the ERP workflows or or how like you
[43:40] know just just give away those open
[43:42] source codes for people to leverage
[43:44] whether it's VBA whether it's appcript
[43:46] or whether it's hands on and so people
[43:49] can start leveraging feel bad I didn't
[43:51] do much this year yet so I will
[43:53] definitely do a little bit more uh you
[43:56] know publications um like basically have
[43:59] LinkedIn articles or blogs or video
[44:01] recordings. Um, you know, cuz last year
[44:04] alone, like, um, I think I published
[44:07] four or five softwares for people to use
[44:10] for free in Yeah, I
[44:13] know. recognition, calculation, all
[44:15] that. I definitely want to do a little
[44:16] bit more this year for that. Um, and and
[44:20] on another side is um, you know, I'm
[44:24] having the SCP has a very good
[44:26] initiative right now um to promote like
[44:30] technology, right? You know, I'm always
[44:31] on the CITP committee to promote the
[44:36] CITP credentials with C IT professional
[44:39] credentials for ASPN. Uh, and this year
[44:42] I'm also trying to see that we can
[44:45] leverage technologies and combine
[44:47] technologies more embedded technology
[44:50] more into the foreign topics as well.
[44:54] Especially at the moment there are
[44:58] there are tens of like like AI is a big
[45:01] risk right itself is big risk uh like
[45:04] the computer vision side like deep fake
[45:07] for example is is always a very huge
[45:10] risk yeah huge yeah there's a cat and
[45:12] mouse thing right like we need to keep
[45:14] up we need to keep up from a regulatory
[45:16] perspective yeah exactly and and that
[45:19] this is one one for a single county uh
[45:22] risk with AI And second way, there's a
[45:27] whole
[45:28] uh cryptocurrency thing right now. Uh
[45:31] like there are just so many fraud going
[45:33] on right now with cryptocurrencies. And
[45:37] I think the the current uh uh political
[45:43] uh environment is
[45:45] really really good for those fraud to
[45:48] grow substantially
[45:50] uh like in the recent years. So, I I
[45:53] think this is where uh you know, hope
[45:57] where like I'm hoping we can use
[45:58] technology to tackle some of the
[46:00] cryptocurrency fraud as well. Um I love
[46:03] that. Yeah. And I completely I'm
[46:05] completely with you. There's
[46:06] definitely there's there's a question of
[46:09] ethics when you have one of the top
[46:12] members of government also own like five
[46:14] companies, five major companies in the
[46:17] US, right? Not knocking Trump there, but
[46:20] come on guys. like and he could shift
[46:21] around IP from one company to another.
[46:23] Yeah. You know, it gets a little a
[46:25] little sketchy pretty fast. Um, man, I
[46:27] we could talk to you forever. We got to
[46:28] we got to wrap it up, but uh I would
[46:30] love to Yeah. Any of that stuff,
[46:32] especially if you're doing trainings or
[46:33] whatnot with the ACPA, I would love to
[46:34] work with you on on some of that. I'll
[46:36] try to migrate more into that education
[46:37] space. Be a lot of fun to work on
[46:39] together. Fosting a couple cat with this
[46:44] year. So I will invite you to one of
[46:46] those fire chat and talk about you know
[46:49] what you have been working on what you
[46:50] haven't seen uh you know in the in you
[46:54] know well you whether you're working at
[46:56] Google syntos or in other or in other
[46:59] spaces you're saying like I I want to
[47:01] hear your
[47:03] yeah I would love to be on the fireside
[47:04] chats and I'd love to get more back into
[47:06] the AICPA so maybe I could chat about it
[47:08] with you but yeah everyone I'll put I'll
[47:10] put the links to all of the stuff you're
[47:12] working on in the chat and we'll
[47:13] definitely talk again when there's more
[47:15] time. I'm sorry we had some tech
[47:16] difficulties at the beginning. So, I
[47:17] appreciate it. Yeah, of course, man.