Demystifying AI for Grant Writers: Burning Questions Answered!

hey there you are in for a treat if you've been  like me and everyone else in the collective   you've been likely thinking how is AI going to  impact the grant writing profession so whether   you're feeling hopeful about it because it's  going to eliminate some of the things that are   really hard like writer's block or if you're  feeling some fear or maybe even a combination   all of that's normal and one of the best ways that  we can start moving through the unknown is to go   talk to an expert directly so for that reason we  have brought in Philip Deng he is the co-founder   of grantable he's been thinking about the  intersection of AI and grant writing honestly for   two years well before ChatGPT hit the mainstream  in December of 2022 right like it he has been   hot on the pursuit of this figuring out what is  this going to look like how do we actually make   a tool that makes current writers lives easier  so you are in for a treat he's super down to   earth I can't wait to introduce you so let's get  right into it and have a fantastic fireside chat I hope you guys brought your candle because we're  calling this a fireside chat so got mine I have to   get the wicked candles because sometimes I forget  they're burning which I know is very dangerous and   if they're wicked I hear them oh sweet what do you  have for a candle what's the smell I don't think I   can identify it but it's nice I've never done this  before so I like having this kind of fireside chat   yeah um I've got a series of questions but I was  hoping you could take a little bit of time just   real quickly to introduce yourself and we'll get  into your background in a moment but if you could   just give a quick intro we'll then jump Jump Right  In yeah um it's it's so cool to be here finally   we've been talking about this for a few months  now so I've really been looking forward to it   um and this is an incredible turnout so I'm  really honored to be here I'm Philip dang I'm   the co-founder and CEO of grantable my  background is mostly in the non-profit   sector so for about 15 years I worked all over  the world in the Marshall Islands in China   um as a front line service provider as a program  manager and I started a non-profit in Seattle   but I ran for five years and then I ran a grant  seeking program for a big land trust and then was   a freelance consultant also um all of that before  starting grantable a couple years ago exactly I   love that you've actually touched grant writing  from all different angles because I think that's   helping with actually building out your product so  let's just jump right into the juiciest question   everyone wants to know is AI going to take away  my job as a grant writer can we get into that   you can definitely get into that um I increasingly  I think no um I think we were all sort of   um stunned when we saw what chat GPT could do  initially back in December when you were bringing   our memories back there but you know I think what  it's going to do is certainly it's going to change   the grant sector um in a few ways one I do think  that the amount of time people spend on the actual   assembling of narrative um character limits all  that kind of stuff the assembly process I think   will become less intensive it will be faster  with AI um I think the types of information   that we can include in Grants will also start  to enrich I think research is going to get   um more more powerful in in individuals research  capacity is going to increase we should talk about   that um at some point today though because I don't  think chat GPT is uh a good research tool yet   um so that's a Nuance to dive into um and then  lastly I think the the number of organizations   that go after grants is going to increase and  that's really what I think will be the main   driver that actually increases the amount of work  to be done for the sector you're saying number of   organizations going after grants you anticipate  will increase why because the barriers to actually   putting together a proposal are lowering yeah  well I mean the barriers are one of them is   sort of the technical know-how the capacity to  put that together but I think in addition just   the world becoming more technologically connected  and the sector itself really prioritizing Equity   um some of the stats that I've seen are that one  percent of applicants take down half of all grant   funding every year um 93 of organizations that  apply for Grants do so uh without external help so   seven percent are employing external Consultants  or Freelancers to or firms to to work with so   um and then of course in in the sector at  large I think 75 percent of organizations   um have a budget of less than five  hundred thousand dollars a year I think   um something like that um pretty close and so a  lot the the vast majority of organizations out   there are either applying for very few grants  or aren't applying at all or aren't accessing   help aren't accessing opportunities so I think  there's a lot of potential out there um to expand   the community so what do you think from like a  competitiveness perspective however because we we   teach our members like hey we want to be targeting  grants we've got a 20 or better chance of winning   so is this you know I mean that can really change  that Dynamic too of all of a sudden now funders   are inundated with so many requests are they  going to start and they don't necessarily have   other imp Tech infrastructure in place to handle  that still a very human process of reviewing those   grants presently so what changes do you see to  accommodate for that well I do think that they're   going to adopt similar Technologies um you know  in the job application market right now A lot of   people are using AI to write optimal uh resumes  or cover letters and then the companies are using   AI to read them so I don't think that's an ideal  scenario where you just have Bots talking to box   um but I think that's another role um a direction  that I hope uh Grant professionals the the sector   will move to be something akin to like more like  real estate agents where you're actually your   um your insights the things the AI cannot do  but helping um the right applicants find the   right funders um I think is going to be something  really important increasingly important um so I   think better alignment can cut down on the number  of proposals that funders are seeking in a in a   way that's more effective than simply just like  hiding away from the world or using AI to read all   your applications there's a big opportunity that  the funders are interested in as well um to have   the right applicants find them and vice versa for  the applicants to find the right funders and so   still I think as many databases and stuff as there  are out there which I think are a starting place   um I I think the role of of people here in our  sector especially as terms like um trust-based   philanthropy relational uh Grant making are  becoming much more prominent what that generally   means is that human beings should be in closer  connection with one another operating in more   Humane and relational ways so I think that's where  the work should increase that's where I'm hearing   thought leaders um like voule has been advocating  for stuff like that Community Centric fundraising   um those are all the principles that  the sector has been trying to pursue   um and so at this moment it's like can we how  can we use this technology and this moment to   guide the sector in that direction yeah and one  of the things that makes me think of is when   when we first met you were looking at all of the  six different phases of content that we teach in   our program and you said we're going to plug in  to this portion of phase three the writing portion   like it is not this all-encompassing taking over  the research and the project management and all   these other elements right and so I thought  that was just a really good reminder how you   even yourself like spotted this is where we plug  in as a tool not taking over this entire ordeal   um there's a good question I have a ton of  questions but I'm going to um weave in some that   are coming from the chat which I love this one  it's got a lot of Thumbs Up are what will happen   when the grand Tours start using AI detectors  to gauge authenticity can you talk about that   yeah the detectors don't work very well um if you  just look for a sampling of the news stories um   that are covering that subject right now obviously  the place that they're being used mostly is in   education so students that are supposed to be  writing um writing on their own teachers are using   the ChatGPT Checkers and it's kind of going wrong  in a few directions it's it's basically having a   lot of false positives and and missed negatives  so human written content is being flagged as AI   written and I mean this makes sense right because  the AI is trying to create very uh life-like   writing patterns and so you've got an AI kind of  trying to trick an AI so you have a cat and mouse   game right here I think you can keep making the  detectors more sophisticated but then the other   models will get more sophisticated at evading them  so I think really it's it's more like uh we as a   society need to come up with um a more nuanced way  of determining if we want originality as you do in   a classroom setting how do we test for that how  do we assess that and then I've written about in   my in my newsletter that I think in the grants  context I will be advocating for authenticity   um being uh the the value that we prize over  originality I would say that that's already   sort of happening with copying and pasting  if you use Grammarly to update your writing   you know it'll suggest a better phrasing and you  click and it just changes that's essentially the   same technology just in a more limited context and  less flexible than the GPT models but it is still   kind of it's breaking your writing into numbers  into patterns and then suggesting better or   sort of whatever what it thinks are more concise  patterns so um unless you really go for an all-out   ban on spell check Tech Grammarly all that kind of  stuff I think it'd be difficult to draw a line in   in that in in our space I think it's easier to  do it in a creative writing class at a college   where you really are supposed to be growing as a  student in our situation I think you definitely   want to be a hundred percent factual you want the  authentic voice of the client or the applicant   organization to shine through but doesn't have  to be a new proposal every single time or do   we all actually have our Master documents and our  boilerplate and our copying and pasting techniques   um so I think it's a little fuzzier in our space  yeah 100 and I think could be helpful just to do   a really quick over works for everyone that  didn't have a chance to set up a free account   or start using it can you just give like a high  level overview of the functionality of grantable   just so we can con juxtapose out against other  AI tools yeah so if you when you sign up for   grantable you upload previous writing samples  um ideally a proposal that you've just written   um the more the lengthier the proposal the better  um it can always compress it's harder to expand on   things you haven't written um and then uh you can  basically just uh work in our in our workflow and   have it draft proposal or responses to new prompts  that you encounter and the process is that you put   in the prompt from an RFP you give the model some  additional instructions like a character limit or   tone specifications the how you want them the  model to draft it then the final stage before   writing is to select your Source material so you  can either go through your own work and highlight   specific Snippets and say use this use this this  or there is a feature where it will search your   documents semantically and pull out what it thinks  are relevant to answer the prompt and then it'll   give you a draft which you can edit manually as as  we all normally would and then you can also call   the AI back to help you work on any segment of  that output so you can make a particular paragraph   shorter or longer you can change tense all that  kind of stuff that you've probably anybody that's   played on chat gbt kind of has an idea of  what the models do yeah 100 so I think this   is important to clarify like is so grantable  is powered by chat GPT correct or not almost   um grantable is powered by GPT um so GPT is the  large language model developed um the style of   large language model developed by open AI which is  a research lab from the bay area which um probably   a lot of you have heard of now the latest model  they've created is GPT4 and chat GP T is it's more   of a user interface it's a way that they created  a a way for the the public in general to interact   with the language model and they actually released  it as like a research trial and they didn't think   it was going to take off um because it was in  November it was still powered with GPT 3 or 3.5   and GPT3 actually came out in late 2020 so it was  sitting around for quite a while and people just   didn't didn't have that experience and so the chat  experience I think was what made the understanding   uh visible to the to the whole world yeah this  is helpful so I think like one thing that people   are are wondering about is is okay here comes  the tough questions you ready for it because my   audience just has got lots of them so if you're  ready we're gonna dive in so they want to know   how is grantable better than ChatGPT for a grant  Raider okay I'll give a little bit of context if   you want which is sometimes for especially  for because we serve a lot of those that   are freelancing and Consulting sometimes you're  dealing with like it's the first grant to go out   for an artist let's say or it's they don't have  this huge repository of information to call upon   and so that's where I think sometimes there's a  little bit of a rub versus if you're within an   organization and you have a ton of information  to go dump in there so can you just speak to   maybe like that scenario but then also just just  how those two little beasts compare yeah the the   first scenario where you're starting off with a  client for the first time and you're it's really   Step Zero um I actually I want to highlight that  moment because I think it's another time where   the skills of a well-trained grant professional  absolutely shine it is that is the mo that is   maybe an even more important moment now when  you have the ability of AI to uh perpetuate   whatever you're creating in that in that first  go so I would say you know if I were coaching   um folks in in a grant writing firm I would say  like let's use GPT uh for scaffolding let's use   it for ideas but don't um you should have  the finals say you should really be the one   that is putting your stamp your signature  on the work at the end and feeling that   sense of ownership and pride over the craft  from there I think it's great to use AI even   more and more especially when there really is  no new creative work that's being called for   um so that's how I would use either grantable  or you could use chat GPT for that as well   um if you're talking about ideation or  um that that sort of pre-writing Step   um the way that I've explained the difference  between grantable and chat GPT is like to to say   uh chat GPT is like going to a restaurant whereas  grantable is like having a personal chef come to   your home so if you go to a restaurant and you  order a dish they're going to use the food that's   in their kitchens in their pantries if you have  a chef come to your house you're saying I'd like   this dish but here use my food Use the ingredients  I purchased I know what's going in and so then I   know what's coming out as well so you could get  the same dish from both places but the one that   is your own your grantable one is going to be much  more personalized to you and assuming you know the   chef and you develop a relationship it kind of  it will stay and have this growing understanding   of the work with a particular client yeah and I  think this segues into one of the reasons people   sometimes get like we're trying to start tinkering  with AI and using it and exploring what it means   so maybe someone goes into chat GPT and they  use it and then they're thinking like wow this   writing is not me or this is so it obviously it's  excessively polite but sometimes you're like oh   that's like not that's not my tone that's not my  voice the things that make writing distinctly our   own and so obviously that's a one reason grantable  is highly useful is it starts to learn your voice   but can you just talk about how different it is  to learn AI than it is to learn how to use a cell   phone or perhaps like a piece of software like  it's a different Beast to learn and I think that's   something that we're all struggling with because  it's been a while since maybe search maybe when   Google came out would be like the last time we had  to learn how to use something in a very new way   I think that is an incredibly perceptive question  and I would say that if I'm at all further on than   the folks in this in this webinar it's not by much  um I've you know I'm 37 I've you know Googled and   used smartphones since you know roughly College  days so you know I have these habits ingrained   as well but I agree with you this is a very new  way of uh sort of retraining my thinking when I'm   working with an AI software I think it's training  me to be more precise um there are all kinds of   guides out there that uh people are starting to  write about uh tricks for Effective prompting   um but some of the best I think are specificity  uh closed-ended uh tasks if possible so you know   you give it boundaries um if you can cite other  examples that it can replicate that's helpful uh   as far but you know I would say the thing that I'm  I've been talking about the most is to understand   where these models are extremely impressive and  and intelligent in a certain spec respect and   then from a different angle they're com they're  very very brittle and I think the way to the way   that helps me remember how what I'm interacting  with is to understand how it's working which   is that a huge amount of training data was put  into this model and it was analyzed under like   an unprecedented amount of computing power and  number of parameters so hundreds and hundreds   of billions of data points for all of these uh  pieces of text that are mapped together and their   probabilities and linkages are all calculated  and this unbelievably big equation which is why   it costs so much computing power to to do these  calculations and from that these language models   are deriving the ability to to predict what  should come next in a sequence so when you put   in a prompt it's actually generating its output  based on probability it's not looking through   its training data and saying this looks like the  thing that the person wants copy and paste it's   like it it has no concept of any of that stuff  all it knows is each next part of the sequence   that is most probably supposed to come up so if  you choose something that is very easy if you   say Mary had a little lamb it will probably write  its fleece as white as snow because that phrase   has appeared so much in its training data it says  okay it's probably this or a really interesting   uh illustration would be math arithmetic if you  do four plus four it'll probably say eight and   that's not because it's doing math it's because  the text four plus four if equals eight occurs a   lot just out in the wild in writing but if you  change that to two four digit numbers that are   much less likely to appear in the training data  at significant rates it'll just guess that oh I   know a number is supposed to go here and it'll  spin up a completely you know incorrect answer   so I think that illustration really shows how  these models are working and if you can start   to remember what's happening kind of mechanically  on the back end you can start to say if I'm asking   it to do something like change the tone well it's  got a ton of material that it's studied where it   extracts tone so it's probably really good at  that but if you ask it to do research which a   lot of people are doing or asking it facts about  the world some very general facts it'll get right   but other times it'll make up entirely fictitious  organizations it's just doing it's creating a like   a prop almost like a movie prop and it's just  supposed to look like a correct answer so if you   understand how these models are working I think it  helps you to to start to know when can you push it   and keep asking it for something better develop  your prompting and then when should you not even   really uh ask the model for um or its assistance  I guess 100 I think this is why I love your   newsletter and your podcast so much the process  will definitely make sure we make that available   for everyone later because you're doing a very  good job of explaining what's actually going   on in the back end like I uh you know briefly  learned how to code so I could code our website   but I never actually understood it it still was  always just felt like it was magic happening and   what's cool about what's going on here is that  this actually we can't understand happening   like it's I mean it's still a monolithic thing to  get our arms around but it's um it once we you're   right we understand its limitations and how it  fundamentally is working and we're on the earth   we're in the early days like it's making sense to  us now versus dropping into I don't know 10 years   uh skipping the next 10 years of life and then  dropping into Earth boy that would really suck   it'll be too much of a culture shock so I  guess this um one of the next questions I'd   want to Tinker with would be so that we've got  a uh someone who brought up that her husband   um and their friend Tinkers with AI prompts every  day to get art and Graphics like pull you know   using the right keywords to kind of bump their  artwork to the next level so can you speak to just   like the educational component of how we start  really learning how to use AI in in a writing uh   appropriate way like to your point the specificity  of close and tasks like that all isn't learned   skill so can you talk about like your plan for  addressing the educational side of using your tool   well so my my partner is a Learning and  Development expert so I I I've raised   my bar in uh assuming what I know  how to teach especially so I'm not   um I I actually my first job was a teacher  but I'm not an educator by training   um so my plan is to learn I wrote about this in  the process a while back my basketball coach when   I was a kid uh taught us an acronym beef balance  eyes elbow follow through or how to shoot a great   set shot and we just practiced it I've I've I  don't know tens of thousands of times by now   um and he also said practice doesn't make perfect  it makes permanent um meaning you know really uh   what you do is going to form your habits and  the final thing that he said was aim small and   Miss small and so he said imagine there was a  dime balancing on the front of the rim and you   want to gently brush it off with the ball when  you shoot and his point there was if you're if   you have Precision if you aim for precision  even when you miss you're going to miss by   less so I've actually tried to take my basketball  training and adapt it to my AI practice so I am   eager to try new tools I try to have fun when I'm  doing it so whether it's image stuff text stuff   I go in I I learn about what it is I read  articles on it I listen to podcasts about   the technology I listen to discussions I listen to  the testimony last week before a senate committee   or Sam Altman uh and uh two other panelists Sam  Altman's the CEO of openAi um testified before   Senators about AI so I'm educating myself  I'm trying it a lot I'm trying to have a   very specific idea of why I'm doing anything  uh when I use AI in my work and in my life   um and then I'm trying to have a very balanced  approach where I'm not all in on it I I don't   think it's a Panacea and I also don't think  it's you know it's not doomsday either I think   I'm trying to have a healthy balance of curiosity  and uh healthy concern just about the power that   the tool represents um so I think chat EBT  is a really good place to start I mean it's   to your earlier point about learning how to  interact with software in a new way it's wild   that you can talk to chat GPT and ask it how  to use it and ask it for tips on how to prompt   um so that's something that I think is a technique  where you can actually ask the tool to help you   use it better which is not something that's prior  been a very useful thing to do but um yeah you   can you can write a prompt see what happens you  could you could have an image generator open and   chat GPT and you could prompt the image generator  to generate something see what's not right take   your prompt go to chat gbt and tell chat GPT hey  I used this as an image prompt and I didn't like   this how would you make it better and then chat  GPT will improve your prompt which you can put   back into the image generator and see what happens  so I think it's exercises like those where you're   coming at it from all different angles with a kind  of a childlike sense of um like the learner's mind   The Beginner's mindset um I think we're all there  right now so hopefully it's not too hard for you   to access that I know I am just sort of uh you  know amazed and unsettled unnerved all the time   in this space so that's that's honestly I love  that this is your approach I hope everyone else   is like feeling I don't know some appreciation  that Phillips not just like a Die Hard can't   see any other way AI or die you know I just  feel like because you have this healthy dose   of yeah curiosity balance with concern um  it it actually I feel like both better for   trusting that you're going to be using this tool  really thoughtfully and so I really appreciate   that approach that you bring to it and it's very  apparent in the process podcast and newsletter   um also just throwing it out there maybe we should  be your education partner because we do teaching   all the time so we'll maybe team up with you  and Robert and help uh help with the education   side huh yeah I I am not trying to be an expert in  things that I'm not an expert in um ask me about   remodeling RVs and you know we've got a different  different story but that's how I feel about the   like the text specifically I'm like thank goodness  we met you okay let's partner up let's do it so   one of the concerns that comes up often is around  privacy of information from clients so can you   just address that in terms of like obviously some  information like plugging in an annual report is   public information but some people are concerned  particularly with complex grants that are using   you know they had to sign an NDA to work on it  how is that information placed into grantable and   protected if you will well so the information you  put into grantable that's stored in your grantable   account it's uh it's a state-of-the-art backend  called superbase it uses you know all the industry   standards security it's not something that's  these these kinds of components are more and more   um sort of ubiquitously secure they use the  same components AWS back end that's Amazon   uh web services um so I mean the biggest firms in  the world are what were built on top of as far as   just holding your account data I think what most  people mean though when they ask about privacy is   what happens with the AI model so you can read  openAi's API policy so the API is how other   platforms communicate with the GPT language models  so according to their policy on their website they   hold the data that's sent to them for 30 days days  they say they're checking for malicious content or   Bad actors after which they delete it and they  do not train on your data unless you opt in and   want them to want to participate in that in that  way um and then they're not apparently opted in   do you have to opt out or you have to opt in the  way that they've done it is it's an opt through   the API it's an opt-in and I I think I heard  uh Altman testify that for chat it's an opt out   um but maybe I misheard him in the testimony yeah  no that's super interesting and it's linked up in   the chat box so thank you for someone for doing  that is grantable going to be able to do anything   with Grant reviews and Angela you asked this  question so if it needs clarified please do it   so let's say you have a 150 page proposal and you  wanted to give you a quick gist of what it was   before you dove into it I tried using chatGPT and  it blocked it because it was 150 pages and even   with shorter proposals because there's formatting  in the PDFs you convert it to word and then paste   it in and it kind of freaks out so I was just  trying to get summaries of ones I didn't get   a chance to reach we are working on that um I'm  happy to talk about where grantable is gonna where   we're headed in the in the next phase here but  you're running into um basically infrastructure   level problems so there is a limit of 4 000  tokens which tokens are actually fragments of   words so that's on the language model's end if you  think about how many hundreds of billions of data   points there are in these in these calculations  you can see why at 4 000 tokens or word fragments   the number of points that the models are handling  at that point is is incredible it's huge so they   have to limit the amount both in and out so all  in it's 8 000 is the limit right now and that will   probably change as the models get better there's  more computational power that they have access to   you but that's a hard limit just in terms of  computation so it's really difficult to throw in a   whole document just because it exceeds the length  limit so you could section it out um that's that's   one thing PDFs are just a really wild file type so  they they're they're messy I mean I think they can   look good in the way that they're intended but  if you ever have to uh move them into word or   anything as we all know they can just get really  really wonky very quickly and That Remains the   Same when you try to extract the the information  out of them they're actually really pretty messy   um at the level of how they're constructed um so  that's something we're trying to work on there are   different uh tools we think we have a pretty good  one in grantable that will if you upload a PDF   it gets it pretty cleaned up just to the text  that you want um but where grantable is going   is we actually are beginning to talk to a few  funders out there and we're trying to figure out   um from the other side of the problem uh  in terms of review in terms of alignment   um all the grantors out there that do all all the  information sessions possible to try to educate   prospective applicants to make sure that they are  really in alignment and that this is the right   opportunity we're working with them to try to  come at this from the other direction now and see   how some of the same features and tools that we're  building can meet so it really is like um building   the bridge from both sides and I think when they  do meet we'll end up with a really really cool   experience where or the applicant I think we're  trying to go towards something where it feels   like you have a representative or an expert guide  from the grantor that is helping to advise you   as you're working on a proposal and then for The  Grand Tour this guide is somebody that should help   um to align their proposals much more effectively  to sort of discourage or at least caution people   that are really out of scope um and in the end  perhaps also try to to work on a little bit of   um uh evaluation as well but I think we need  to proceed pretty carefully there otherwise you   get the robots talking to robots scenario all  right so this is a demo account I took real uh   Grant applications and anonymize them and then  um so basically you upload a Grant application   um so this was a docx that kind of lays out  a bunch of information about programs and and   Etc and then uh this part is just like your Google  Drive you can create a folder so we'll call this   GW demo um and then when you want to create an  answer or content you create content and then   you can respond to a prompt we're coming up with  some other stuff but um you could copy uh you know   a question from a a grant proposal so like uh  please describe your programs or whatever it is   um and then you can say 100 word limit or  whatever um continue then you have the option   to search uh manually so you can search through  all of your previous writing you could do uh   you could search program and then uh find wherever  the the program information is so this is a   here's a nice organizational overview so I could  copy and all you do is highlight and press Ctrl   C and that throws this snippet in here so that's  the manual process or you can do auto search and   it will just find a bunch of stuff that it  thinks are useful um and then you see your   input limit here that's the tokens that I was  talking about and then we'll just see what it this makes me want to ask about formatting because  that's been an issue too like you try to throw in   bullets and then the bolts go away so like is this  able to add any of that like functionality Beyond   paragraph text we're working on that um the the  way the feedback that we've gotten so far is that   um when people format they generally go to word  or Google Docs where you know the the more visual   adjustments can be made Um this can handle it it's  just that sometimes then it creates other problems   so in prioritizing what we're doing at grantable  I only have we only have my my co-founder Robert   is the only engineer on our staff right now so uh  it has to be sort of we have to manage what we do   otherwise if we enable a feature and it causes  other things to break or more support tickets   so right now it's sort of like in a waiting  pattern but we're looking at bullet points you   know graphs images all that tables all that kind  of stuff coming coming soon so right now this is   how it generates um and then you can highlight  different sections and you can say condense   or something and it will condense that section  you can have it change the tone or whatever or   um you know edit this manually um however you  like you can you know undo and then uh there's a   response history that will track as you're working  through so this just condensed all that so you   can replace it and cut it just about in half but  if you decide oh I don't like that then you can   restore this other version and so then basically  we advise once you've workshopped it to the point   where you like you can either paste it into the  portal directly or you can bring it to word or   Google Docs however you want to to work uh with  with your colleagues or with your clients but the   one thing we say is if if the final version is in  grantable it's best because because then it has it   for the next time you want it to answer something  and it will prioritize your freshest uh writing   yeah so how are you seeing people use this  workflow-wise like how quickly are they then   like they're doing the ideation I think this is  like what everyone I think needs to take away   from from this is like this is what like helps  with that messy part of like just getting text on   the page in the first place working through your  first second maybe third draft but then okay now   we need to go get it like in a Google doc format  where we can share it with others so like how   are you how are you seeing people make that leap  like yes I think so I mean it's quite a range I   mean some people some people honestly just kind  of use it as is they don't do a lot of shaping   but I think it's everything from just ideas to  and then doing a bunch of revision with teams   um we're as you can see working  on bringing collaboration in and   actually our next project is to work  on data um and different quantitative   elements that you'd want to stay consistent  across uh reports or proposals so like KPIs,   budgets and stuff that never that you know stays  the same for quite a while these data points   um so we're working on bringing each part of the  grant workflow into grantable in a systematic way   um so that's sort of where we're headed next  that's that's how we're seeing people use it   right now is in this flow it actually took a  lot of iteration to get to this model and what   we realized is one of the things that we're all  learning with AI right now now is how to work   with it and so one thing we found is that these  early steps in in grantable at least these are   your big big rough changes so if you change the  prompt it's going to change the entire output of   course so these are sort of the most Upstream  changes then each successive phase you're fine   tuning more and more until you finally get to this  place where you're actually like literally editing   single words or letters and that we've found  is a really important um we think it's sort of   a principle that's emerging where you have the AI  move you through the really the course broad steps   and then bring you to the fine tuning phase much  more quickly than if the human were doing every   single one of those phases alone but what I think  gets messy is when you Mash all of those phases   together like you might have in chat GPT where it  can be difficult to adjust an upstream variable   um after the fact so the reason we created it like  this is because if you get to three and you don't   like it you know you can take away um you know  some of these ingredients you can change the   prompt you can change your direction you can go  back and forth basically zooming in and out and   it will redo what it's doing based on your changes  so it's really about creating the controls and the   fine-tuning knobs for people to be able to use  the AI effectively so I'd say as simple as that   looks it took so much work to get to this uh to  this point and I think we're I think we're on the   Leading Edge of this I think people were going  to start seeing some really really elegant ways   to work with large language models and I think  there will be some principle here about like it's   almost like a transmission honestly it's like low  gear and high gear there's something about that   type of understanding or I I talk about it like  a knife sharpener that I have I love to cook so   there's like the coarse wheel the medium wheel  and then the fine polish I think something very   similar that is possible with AI now at least in  in the writing context where you should know which   wheel it is so do you need to make huge changes  or are you trying to fine-tune it's not just   a one-size-fits all yeah and I don't know about  the folks on the line but I always find like the   hardest part is the early part it's like getting  those big chunks down like that's the part that   can be really tough so one question I'm having  is folks want to know about collaboration is   that already set up like should they be bringing  in their client to work within there and I could   say some pros and cons to either case like this  could be your messy work Zone yeah yeah I rather   only have them operate out of a Google doc where I  can kind of control their attention but I can you   know maybe have a tech Progressive client you want  them in there can you talk about collaboration yep   so depending on what you mean by collaboration um  you can have multiple people come in and use your   account right now and the situation that is  not yet handled is literally the simultaneous   editing where like if you've ever been in a  document and you see the other person deleting   stuff you know and you're both working in in real  time together that's actually pretty nifty code   um and you can imagine like when two people make  a you know a conflicting edit at the same time how   do you resolve that it's actually sort of a it's  well it's not a fun problem if you ask Robert but   um other other than that situation um granted you  can multiple people can work in the platform and   they can you know you can work together that way  but I think Meredith raised an excellent point   like do you want that to be in word or Google Docs  we are going to bring in chat we're also thinking   about some cool um like you know how when you need  data from somebody on the team that's really only   just needs to give you a couple numbers so we're  we're thinking about ways where we can actually   create a process where you come to the point where  you need the data and you can actually send them   a request which they'll get by email and they can  either respond to the email or click on a link and   they'll interact with grantable and grantable will  ask them for the information they'll put that in   and then it'll go into the place where you need it  so that's great because I know there's been times   where I'll end up having to take out of Google  doc like just the section make a whole other page   for them so they won't go get sidetracked but just  answer this one paragraph so yeah that would solve   a big problem yeah so we're trying to do that  and then also those those data points usually   show up again and again and we lose sleep  over their consistency so we're working on   all of that as well so that you can say this  year's annual operating budget is this the   program budget is this we helped this many  people at this event and those consistent   um across your different work so so just so we  can like round robin some of the really quick   questions that have come in that are a little  bit more feature focused can you talk about   the token system people are confused by that  yeah so the tokens um thanks for your patience   with that that terminology is coming from the  tokenization of words so I talked about that   earlier the words when they go into these models  actually get broken into their component parts   and the process is called tokenization and  so um basically when people are talking about   interacting with these models in text they're  measuring the amount of text that's going in   and they're measuring it by tokens and  so that's sort of become the the language   um put out by the model makers that we've we're  using and then what we're trying to do is create   that consistency through to the the users on  our platform as well so they're not like crypto   they're not anything like that it's sort of like  fuel in your tank or your battery it's like how   much AI usage do you have left the reason that  we're not doing it straight to words is because   eventually there will be really cool features  that will still use uh tokens or use credits   we could change it to something like credits  but that won't actually result in a in a word   or a written result per se for example like if  you're using a researcher to ask for Census Data   um that will probably use a bunch of AI power  computing power but you know emotionally generate   you know a little result for you and it so it we  thought about going with words but um we're trying   to communicate the concept of AI usage yeah 100  yeah I like the battery analogy that's a solid one   uh just for our photo unicorns that live abroad  do you does grantable work in other languages   yes um specifically yep German yeah yeah um it's  getting better than your state open AI is putting   out those updates pretty regularly so most major  Global languages are are pretty good now um I'm   not a German speaker so any German speakers out  there will have to judge for themselves but some   as I hear it's pretty cool that's wonderful um  Mira there you have it okay so one of the things   I think people are confused about is like you  mentioned PDFs are a real Beast so like how do you   use PDFs and get that information in a grantable  like are you actually just having to go highlight   all that text but then like the formatting goes to  hex so what are you suggesting there you can up I   would just say try it um we we worked on this  problem last week actually so we've found what   we think is a really good uh translation method  um that uh we actually we went out and like paid   paid for this so it's not an open source thing  it's it's supposedly one of the better things   um you can use I see a PDF to Word doc converter  I mean essentially it's a very similar thing   we're just we're going even more stripped down  to something called markdown so we're really   trying to pull away all of the formatting even  word has a lot of that too Word files are way   better than PDF files for getting to the text um  but yes if all else fails uh you can you can copy   and paste if you wanted that excerpt you can  paste it into this box here and then it will   be part of your library but give it a try just  try uploading it and seeing what it looks like   um so you it would show up here and then you  can just click it open and see if it's all all   mangled but hopefully not cool well we're coming  up on the top of the hour is there something that   you wish I'd asked that I didn't get us into  so something I've been thinking a lot about is   when the AI optimists that like like Sam Altman  and others talk about this future that you know   is is the preferred future the AI future where  it creates a ton of value which is distributed   and frees us up to do uh to have better jobs and  to have more time to spend as we like it's funny   that they always talk about freeing up time to  care for one another to pursue the Arts to take   care of the environment to they they literally  list basically the work of the non-profit sector   and when I heard that I that immediately caught  my ear and I think there's that's no accident   um if you think about the the for-profit the  private sector everybody there is their main   mission is to make a profit and this AI tool is  going to come along and be able to do more and   more human jobs or replace more and more labor  and those companies are all incentivized to do   that because it'll lower their costs and increase  their profit so if if people are shifting if we   want to get to this better version of the future  my evolving hypothesis is we're actually trying to   get a lot of people to come into the non-profit  sector and the Very fact that it is non-profit   that there isn't that motive to replace people  with AI labor and that we are focused on work that   traditionally is not accounted for by the the free  market by capitalism I think that actually creates   a kind of resilience to AI it it it doesn't want  to eat the the non-profit work as much I would say   um and so I actually think there's  this really interesting intellectual   space where I'm calling on the sector to  say hey how what would it look like if   The Nature Conservancy was the largest employer  in the U.S instead of Walmart like we have to   kind of think forward in those types of ways  and what would it take yeah it's an abstract at   first I was like I don't know this definitely is  a Fireside conference I'm like come on now we need   to actually be around a fire with some whiskey to  have this debate yeah but it's because sometimes I   think there's also a benefit like the lines are  blurring between you know for-profit businesses   that have a social Enterprise mission non-profits  that are utilizing Tech more effectively than ever   like it's getting really interestingly a lot more  blurry I really think we're we're at this moment   where I think the non-profit sector which has  been mine for most of my career um and has always   just sort of been like junior varsity you know  compared to the private sector I I don't think so   I think we've been innovating I think we've been  doing business on hard mode like not only do you   have to do the service and provide the service you  actually have to run a whole other business to pay   for it which is like you know it's it's running  two businesses in one and I think we've been doing   that because we're Mission motivated because we  see this work that isn't going to pay but needs   to get done so how do we motivate ourselves how do  we organize how do we find money and resources how   do we use Tech so I think the sector could have  um a moment um a really really important one in   the context of AI in that if you think about  the all the ideal scenarios where people are   doing these human and planet-centered works the  groups that are already doing that work would be   the ideal place to absorb these people and to  bring them into and expand the sector um so I   think I don't know how to get there I'm grantable  is my little way of trying to help that go forward   um but I think that's one way where I hope  the non-profit sector will come into this   particular Tech Revolution with maybe more  confidence than we would have in the past that   is that this technology artificial intelligence  requires human-centered alignment more than any   other technology ever it's not even close and  the the sector that is innovated the most on   human-centered thinking and alignment and values  and systems and organizing planet-centered as   well is the non-profit Center so the intellectual  property necessary to make AI successful for the   world I think lives in the non-profit sector  and you're only proof that you need is that   open AI started as a non-profit for a reason  they saw the problematic potential of having   this be totally profit motivated and there's a  model right now between you know the non-profit   uh people that want it to be values driven and  then the people that are pulling it towards   um just making more money um so I think we have  this really unprecedented role to play in this   technical Revolution 100 no we're tracking with  you on that and this is going to be I think you   just published some content on this so everybody  I want to recommend go find Philip on LinkedIn get   onto his podcast his private podcast it's amazing  this is like he's my go-to for be tracking what's   happening in the industry all right I'm gonna try  to sneak in one more question and then we need to   wrap up I was supposed to remind you ask me about  research capacity verse chat jpt I don't know   exactly what we were talking about does that jog  your memory yeah I do um so we we touched on this   and thanks for the reminder so a lot of people  use chatGPT for research and then say Oh it's a   terrible researcher so it is uh for the reason  that we talked about earlier about the results   are just probabilistic there there's no it does  not understand the meaning it does not understand   the veracity or the accuracy of its products just  that it has put every next token out based on a   formula so when when people ask it a research  question it will respond like a researcher   if you say answer me in the in the style of a  PhD dissertation it will sound authoritatively   wrong and so when I say right now please avoid  using chat GPT as a research tool the reason is   if it's very general knowledge it'll probably  get it right but you really don't know there's   no clear line when you're crossing over into  statistics or data that is just totally factually   fiction um so just because it seems like it's  giving you a quick answer it's such an illusion   and I understand why people are because some  things are right but you don't know what's wrong   um the caveat is that they are working on  combining large language models with search   databases so you'll see Google and bing going  full speed at trying to win this yeah where   you can interact with a search engine as you  would uh they're gonna go pulse census data   and integrate it and all of that right yeah but I  think it's still there's so there's so many hard   problems there in trying to distill down the right  information from huge data sets like the size of   the internet and figuring out what is what is  right especially when right is maybe a little   more subjective so I would just really caution  people to um try to follow your your traditional   research practices once you have the data once  you have the numbers and statistics put those   into chatGPT or grantable and it will keep those  and help you work with them in in different ways   yeah I just love this I think this really buttons  up like what's so important about like that Nation   potentially between chatGPT and grantable is like  as you're building this knowledge as especially   you're doing the research like it is housed in one  place that will continue to serve you and one of   the things when I'm in a private Mastermind and we  were talking about how you know there's an upfront   investment period like you need to put in a couple  hours of giving it good information to have like   your outputs are your inputs and so I think what  we can all take away from this is like you're on   the front end of this you're not getting left  behind have curiosity go Tinker experiment like   go back to your child like state where you're  literally just building like a mud hut in the   backyard and you're having like you're having play  and you're not we're not being so staunchy about   it so in that and if we can bring that energy and  approach to this I think we're all gonna have a   lot more fun with it and um so yeah I think you've  really set us up Philip for like knowing where to   go tinker and experiment more give you feedback  you're building out your product and um and just   kind of relax on this whole thing so we don't we  don't freak out like it's got a long ways to go   even though it's come a long way yeah I'm I'm not  going to try to improve upon what you just said so I appreciate that you've been doing that  thank you so very much and with that you   know where to go find Philip to keep following  him and we will send out the replay in addition   to the very generous 100 000 tokens that  grantable has given us to give to you so   we'll share information on that in tomorrow's  email I think that's a wrap folks people all right

Channel: Learn Grant Writing

Views: 7,059

Keywords: freelance grant writing, fundraising, fundraising ideas, grant writing, grant writing 101, grant writing classes, grant writing coach, grant writing for beginners, grant writing for dummies, grant writing for nonprofits, grant writing jobs, grant writing tips, grant writing training, grants, grants for nonprofit organizations, how to write a grant, learn grant writing, meredith noble, nonprofit grants, online grant writing classes, work at home, write a grant

Id: anBSKk-e8vI

Length: 61min 9sec (3669 seconds)

Published: Wed May 24 2023