Agree with this with one exception: When you’re in the crawl phase there will ALREADY be people in your org who have been playing around with AI - in their own time. Find them, make them ambassadors, give them some time and budget and see what happens.
Excellent framework, David. Thanks also for the nostalgia hit of those early days of just futzing about with the web. Great times, fantastic analogy. We are somewhere between walk and run with my team now. The rest of the organisation largely hasn't started crawling yet.
That’s a great description of a journey that resonates with me since it’s been happening in my company over the last couple of years. The only thing I’d add is that for a larger organisation, there will be multiple, parallel instances of this happening. Not all of them will be in sync though and we need to recognise that some functions or departments will still be at the crawl stage while others are further along.
Hi David, This overview is great. I‘m at the curiosity phase myself, and trying to encourage curiosity among others at the large public, ethnological museum where I work. Ultimately the ROI for us will not be a return in cash terms but a return in the form of raised public and community interest in activities and resources of the museum. Our founding director (Tadao Umesao) in the 1970s already saw the museum as being part of a knowledge economy, and was keenly interested in IT… but didn‘t live to see the emergence of LLMs, or even machine translation.
I work for a big private bank as P.O. of our on-prem AI platform and our biggest challenge is legal & compliance! Making them understand what LLMs are and that giving emps creative freedom is actually a good thing. Nope…not happening 😅
@@Secludy Not necessarily, much more they fear „lazy users“ who summarize e.g. research reports which might contain errors (hallucinations) which would then be forwarded to clients without prior verification.
Mind blowing as this video just described my process during the past 6 months! I’m an IC at “run” phase. Really hope that I can get this pushed through and start seeing things take off. Thanks again for sharing your wisdom
Great video and framework, Dave. In your copious free time, you could practically package this up (expand into a half dozen ‘modules’) and start w a couple use cases/industries that work well as you iron out kinks and probably land a bunch of income. (Better yet, just have an Agent work on it for a bit!)
And you forgot the most important fact right now: Many doesn't even know yet how it works. We are making a science of this to figure out exact how LLMs think. Just a few years ago we didn't even know LLMs had basic vision. To to this day we still haven't learned all that a basic LLM can do. And this so SO dang ccol! yet we do need to slow down a bit.
Yeah great framework. I work in local government in the UK. When I ask people if they've tried using chatgpt or Claude etc, the number 1 answer is 'I don't have time to learn something new'. IT aren't into it, really skeptical. AI feels like a cult lol, you can figure out those who are into it because they usually have a generated image for their teams avatar lol.
This story you wrote might be a good contender for one of the 1st AI generated (high quality, natural looking like with human actors and props) drama series.
I work for a large corporate and they jumped on AI in a big way. Got our own hosted chatGPT instance and it's been trained on all our confluence pages and repos. We can throw in any data and it's all secured in a private instance, I'm very lucky.
What I can tell from my experience is that most companies don't know how to use AI and more importantly, they don't trust it. I see everybody advetising it, but behind that advertisement is inefficient "nonesense". You get information on how AI developed and what the break through was and then you get promppt engineering lessons that are from GPT 3.5 times. And another important point is that I am still consulting Companies on things that could be easily solved by consulting with a GPT. That shows that it is not trusted, that you get the best solution from it. Even though you should at least be able to judge that it is a good solution.
@@mwinsatt Crawl walk run is a very old model for organizational development, and I first heard it as crawl walk run fly from the analyst Jeremiah Owyang relative to enterprise adoption of social media almost 15 years ago. I owe a lot of my career to things I learned from his papers. Brilliant guy. And always ahead of the curve. Guess what he's doing now? He's my favorite AI industry analyst. And I'm still using his concepts like crawl walk run fly and multiple hub and spoke model to introduce and embed AI within my agency. The adoption of AI in a corporate setting is very similar to the adoption of social media, and that's why I'm unsurprised some of the best people deploying AI at companies right now are the same people who deployed social media back in the day. I'd say the secret hack to hiring a CAIO on a budget is to look at people who were the OGs of corporate social media. I know some of them are #opentowork.
AI is great at non fiction and sure does understand whole books. I've been working with several models editing my book for 6 months and it's been a ton of trial and error but I've worked out what exactly I can use it for and how, as well as the limits and how to keep on track.
There biggest probldm is they obssessed with their data. While at the same time sending their data in emails that get processed by the same people providing the models. Which then forces them into not trying or even blocking ai The cognitive dissonance is crazy.
The same cognitive dissonance as generative AI companies banning people for trying to steal their IP, when they are literally scraping every corner of the internet without permission in most cases.
I have to say, where I work we're kind of taking it in stride. We use Copilot behind a firewall, nothing else. No robotics. Our coders are using it, and within limits, it does help.
it‘s counsellor‘s job to tell their client what they want to hear. and, that‘s not easy at all. you‘ll have to read their unconscious just below the surface without upsetting them. 😂
Well it depends on the situation. Things like size, composition, domain, etc. The strategy that is best for one may not work for others. If your company has several tens of thousands of employees world wide, composed of a diverse set of domains, it's likely that you will want to pick a different approach. Companies often stand up a COE to stop the bleeding (i.e. when getting crushed by dysfunction resulting from bad risk management). So, from my perspective, as a solution architect in enterprise multinational companies, it may make sense to stand up a COE or even a federated COE model from the outset, but with the priorities of facilitating, cultivating, and harvesting, not dictating.
I saw productivity boost from day 1. One of the great things with AI is that experiments cost nothing, not even much time. Just tell them what you want and you'll get a result in seconds. Maybe you can try more advanced prompting techniques, but even that is getting less important as models are getting smarter. Only thing to keep in mind is hallucination. Just because the answer looks good, it could be all BS.
You must be half-betazoid, it's like you're reading my mind and making a video to help me. I"m part of the A-Team at my company (purely b/c of my passion for AI), currently working on transitioning from 'crawl' to 'walk'. What is a better strategy; 'Marry' a platform like Slack AI or Google Cloud Platform OR build an 'agnostic' model and use different LLMs & Tools for different biz needs? On the one hand, if you go with a 'Platform' then someone else manages the AI agents, but an 'agnostic' model requires assigning someone internally to manage all aspects of it.
i think the former would make you interate faster. Also i recommend to use n8n for automations + llms. Though what do you mean building an agnostic model? are you training/finetunnign your own llm?
@@kdme thanks for the recommendation (just found out about n8n a few days ago but hadn’t tried it yet). What I meant by agnostic, was maybe using ChatGPT for some teams, Claude for others, Google b/c of some of their integrations. For example, Slack requires the whole company upgrade for their AI features, which is expensive. I think Microsoft CoPilot charges over $300/yr/person. So I’m trying to find the best pricing strategy for a company with over 1000 people.
Of course we didn't learn anything about how incompetent employees can be when it comes to using corporate data in isolation (spreadsheets). I fought this battle for years but ultimately gave up when I realized we can't even figure out the paperless office. Top down AI adoption is a much smarter approach, assuming the leadership isn't as dumb as the spreadsheet jockeys.
As a consultant in AI you make a bad mistake. They are asking you about AI, they are not asking you about their business. So, you don't advice them about their business, you advise about AI. Later, it's THEM that will decide the cost - revenue of a particular AI implementation.
@@danielbistman3184 well, first of all, I think he knows his clients much better than you. secondly, the companies are interested in AI especially to grow their business so in this sense AI is fundamental to build a business edge against competitors therefore by definition AI consultancy for corporate clients is tied to the business.
Can you share which report are you referring to, when you say CoE is not better then Decentralized approach? I like that comment, I would like to validate it.
I'm safetiist ("doomer") but all David said today, I agree. Very valuable thoughts. P doom is not everything, and David is still the best person ti think though all these and the best speaker - influencer. I can see myself doing this acceleration at work and go to a pause protest after. I want all to do safety, not just me. If I pause alone, I'm left behind.
Artists: Artists will soon train on your image and style, change it slightly, then generate far better videos for themselves than you could ever make Dave. 🤣 Imagine the irony. Monetization just won't be possible soon. So, it sounds like you hired an over-priced illustrator since quality graphics gigs go for 300 - 500. And ya typically get closer to what yer looking for, rather than settling for over-baked AI images who's attribution is quite sketchy, and consumer resentment is rampant. There are a lot of published genres that wouldn't touch gen AI coz of the consumer blowback. Anyone familiar with the creative arts knows this. Fortunately, you are in an industry that can fully exploit it. At least until much of the industry labor doesn't exist anymore because they have been replaced by gen AI agents. You know it's coming. 😁
Random question; i'm currently reading a lot of books about AI, but most of them are already outdated (for example; Melanie Mitchell - A guide for thinking humans). I can't seem to find good books about AI published this year or late last year. Can someone help me with up to date books about AI? Thank you! :)
Whenever I hear consultants speak, I imagine how a smart person tries to explain obvious concepts to a dumb person with language the dumb person understands and then takes money from the dumb person to make them feel smart too. Of course, neither does the dumb person really understand nor does the consultant make the dumb person smarter. At the end, its just another way for smart people to trick money out of dumb people. I understand, everyone needs to make a living, but boy would the world be a better place if we just leave the dumb people alone and don't waste energy trying to raise that which cannot be raised.