Andrew Bruce Smith’s Post

An observation. One of the most common types of questions I get asked are around an AI's ability to "find" stuff - yet LLMs on their own can't find anything. Here are the kinds of things PR professionals ask about AI: ➡️ How can AI find and analyse media coverage, social media mentions, etc. ➡️ How can AI find and condense what politicians are saying, etc However, unless anyone knows otherwise, AI can't automagically "find" anything i.e. to find online media coverage, you still need a crawler - maybe AI might be used to optimise how a crawler goes about crawling - but ultimately, AI doesn't do the "finding". Something else has to do that. Once you have the raw data/content, then you have something to which AI can be applied. My sense is people often think an AI tool can do everything, including the finding. Which it can't - or at least the core AI model can't do it. It needs to be augmented with other tools to do the finding. Tom Davenport's recent Harvard Business Review article (Is Your Company's Data Ready for Generative AI?) also had some pertinent observations around data for generative AI use: "While 93% of survey respondents agreed that data strategy is critical for getting value from gen AI, 57% said they had made no changes thus far in their organizations' data. Only 37% agreed — and only 11% agreed strongly — that their organizations have the right data foundation for gen AI. In other words, most organizations have a considerable amount of work to do in order to apply genAI broadly in their businesses." It reminded me of a blog post I wrote nearly 17 years ago. https://lnkd.in/ePr9h6dE I'd read Davenport's book Competing On Analytics at the time - here's what I said then: "There's one basic premise to this book – namely that the best-performing companies today have put analytics at the heart of their strategy. No longer confined to specialists or specific departments, analytics permeates every aspect of the organisation. Decisions are made based on analytics rather than intuition. Failure to adopt a strategic approach to analytics will almost certainly result in poorer performance. The key barriers to be overcome are primarily cultural – i.e., intuition still rules the day when it comes to making decisions. Plus, you need decent data to analyse in the first place." To repeat. You need decent data in the first place. AI won't magically provide that decent data. You - or someone - has to get that decent data. Do you think the PR industry generally is good at finding and securing decent data? If not, will this act as another barrier to extracting more useful value from AI? Let me know what you think in the comments below. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpr

"Always invite AI to the table. You should try inviting AI to help you in everything you do, barring legal or ethical barriers." This is Professor Ethan Mollick's first principle of AI usage as outlined in his new book, Co-intelligence. It's well worth reading - a comprehensive analysis that sheds light on the nuances of AI development, its implications for society, and the ethical considerations it demands. Mollick proposes four principles to guide our interaction with AI: inviting AI to the table; being the human in the loop; treating AI like a person (while defining its persona); and assuming the current AI is the worst we'll ever use. These principles serve as a framework for harnessing AI as a collaborative partner rather than a replacement for human intelligence. He covers a lot - but if I had to single one thing out (and most relevant to PR and communications), it'd be about the role of expertise in the age of artificial intelligence (AI). Mollick underscores a transformative shift: while AI's capabilities continue to burgeon, reaching into realms once thought exclusive to human intellect, it paradoxically elevates the value of human expertise. This duality presents both a challenge and an opportunity for professionals across industries. Mollick presents a nuanced view of expertise in the context of AI, highlighting a paradox. On one hand, AI's proficiency in tasks ranging from data analysis to content creation suggests that the need for human expertise in these areas might diminish. On the other, he argues that AI amplifies the value of human expertise, as it requires human judgement to guide, evaluate, and apply its outputs effectively. The journey to becoming an expert in a world interwoven with AI is depicted as both necessary and evolving. Mollick suggests that foundational knowledge and deliberate practice remain critical, but their application takes on new dimensions. As AI takes over routine tasks, the human expert's role shifts towards more complex, creative, and judgement-based activities. This transition underscores the importance of developing a deep, nuanced understanding of one's field, beyond what AI can replicate. Mollick warns of a potential crisis in training the next generation of experts. Traditional pathways to expertise, characterised by gradual learning and mentorship, face disruption as AI automates entry-level tasks. This automation could erode opportunities for novices to gain experience and for mentors to impart wisdom through practise. However, he also sees an opportunity: to redefine the pathway to expertise by leveraging AI as a tool for enhanced learning and exploration. Ultimately, it as an optimistic view of the role of expertise in an AI world Let me know what you think in the comments below. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpr

Want to hear about a free AI tool that helps creative writers - rather than trying to steal their jobs? And no, it isn't ChatGPT. Or any of the other usual AI suspects. Google's TextFX(*) is an innovative AI experiment that offers a suite of ten creative tools designed to enhance and expand the writing process. Developed in collaboration with the artist Lupe Fiasco, TextFX is aimed at writers and other creative wordsmiths, providing them with a platform to experiment with and generate creative text outputs. The ten tools included in TextFX are: ➡️ Simile: Generates similes about a thing or concept. ➡️ Explode: Breaks a word into similar-sounding phrases. ➡️ Unexpect: Makes a scene more unexpected and imaginative. ➡️ Chain: Builds a chain of semantically related items. ➡️ POV: Evaluates a topic through different points of view. ➡️ Alliteration: Curates topic-specific words that start with a chosen letter. ➡️ Acronym: Creates an acronym using the letters of a word. ➡️ Fuse: Finds intersections between two things. ➡️ Scene: Generates sensory details about a scene. ➡️ Unfold: Slots a word into other words or phrases Take tool number one as an example. I asked it for a similie for "writer's block". It came back with 10 suggestions immediately. One of which was, "writer's block is like trying to type with boxing gloves on." (I put that as a prompt into MidJourney and got back the image below to help me illustrate this post). Or number 7 - give it a word and it turns it into an acronym. Here's what it came up with for "Comms" - Consistent, Open, Meaningful, Mutual, and Supportive Any of the other tools can be used in a similar way - to help come up with ideas that are intended as a starting point for providing additional inspiration to the human writer to create their own content. Not let AI take over the whole process. I've certainly found it quite helpful - an example of how AI can augment human creativity, providing a new dimension to the creative writing process. What do you think? Let me know in the comments below. You can try TextFX for yourself here: https://lnkd.in/eY43Pu5X 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpr (*) Before anyone says it, there's nothing stopping you from effectively replicating the features of TextFM by simply giving ChatGPT or any other LLM tool a suitable prompt to do the same. But this spares you the effort.

This is an excellent analysis of Common Crawl datasets which the majority of Large Language Models (LLMs) rely on - and as Stefan Baack says, most people have no idea about them or the role they play. And the downstream effects on LLM output. #publicrelations #pr #ai #aiinpr

Do you know how you do what you do? Being able to accurately describe and explain your workflow is a vital first step in assessing how AI might help your PR and comms work. It sounds easy enough. Examine your workflow and determine to what extent each element of the process can be entirely done by AI; where AI can support or augment your own role; and those aspects of the workflow that can and should remain entirely the responsibility of a human being. But in practice, it isn't quite so simple. Aside from the fact that not everyone will approach the same workflow in the same way, trying to accurately document exactly how you go about doing something may make you realise that it can't all be neatly written down and codified. In short, it's the difference between explicit and tacit knowledge - two types of knowledge that are fundamentally different in their nature, acquisition, and transferability. Explicit knowledge is knowledge that can be readily articulated, codified, formalised, stored, and accessed. It's easily expressed and shared between people, and is often documented in a tangible form such as manuals, documents, procedures, and how-to videos. Examples of explicit knowledge include company policies, process documents, research reports, and the information contained in encyclopaedias and textbooks. And this is exactly the kind of knowledge that lends itself to AI. For example, it's easy enough to create a checklist for how to go about writing a press release. In which case, you could just create a custom GPT press release writer based on these instructions. But what's missing? In short - tacit knowledge. The knowledge a PR professional has gained through personal experience and is often difficult to express or document. It's subjective, cognitive, experiential, and is often not easily codified or communicated. Tacit knowledge includes skills, ideas, and experiences that people have but aren't codified and may not necessarily be easily expressed. This type of knowledge is often gained over time through trial and error, practice, reflection, and observation. Examples of tacit knowledge include riding a bike, playing the piano, driving a car, leadership skills - and, dare I say it, much PR work. So bear this in mind when determining to what extent AI can support your efforts. How much of your work is based on explicit versus tacit knowledge? And how do you convey the value of your tacit PR knowledge? Let me know in the comments below. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpt

AI adoption by PR professionals is highest in Europe. 42pc say they now use it "frequently". But only 24pc of United States practitioners say the same, lagging Europe and APAC. These are just a few of the key insights from a new Cision UK PRWeek UK 2024 Global Comms Report entitled "Elevating & Evolving". 427 senior-level professionals were interviewed from corporations and agencies across the US, Canada, France, Germany, Sweden, UK, Australia, China, Hong Kong, and Singapore. Here are some of the key findings: ➡️ Challenges and Priorities: Top challenges include the inability to measure impact effectively (50%), securing budgets (48%), and aligning with other departments (47%). Analytics and reporting, media outreach, content creation, and measurement are top priorities, indicative of an industry pivot toward data-driven strategic communications. ➡️ C-Suite Expectations: There's an increased expectation from the C-suite for the PR function to influence company performance metrics like revenue, brand value, and customer acquisition (43% up from 34% last year for customer acquisition/engagement; 42% from 33% for driving sales/revenue). Sustainable growth remains a top priority (60% vs. 54% last year). ➡️ Data and Analytics: The report indicates a sharper focus on using data and analytics, with 42% stating they rely on it more now than last year. Companies are seeking real-time data interpretation to inform strategic decisions and crisis management, with 77% of respondents feeling equipped for crisis management. ➡️ Media Strategies: Earned media is valued highly for media strategies, with a slight increase in reliance from 30% to 33% over the past year, while paid media has seen a slight decline (28% to 25%). ➡️ Social Media Platforms: The shifting sands of social media show Instagram and LinkedIn rising in value among PR professionals, while X (recently Twitter) has seen a notable drop from 50% to 36%. ➡️ Visibility and Influence in Communication Departments: Communication leaders are frequently reporting directly to the CEO (41% globally), highlighting the increased visibility and influence of PR within organizations. What do you think of these results? Are you surprised by the fact that US PR professionals are using AI less often than their European and APAC colleagues? Let me know in the comments below. Link to download the full report, also in the comments. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpr

Can large language models like GPT-4 be prompted to enhance the variety and quality of idea generation - especially compared to humans? A new report titled "Prompting Diverse Ideas: Increasing AI Idea Variance" and authored by Lennart Meincke, Professor Ethan Mollick, and Christian Terwiesch from The Wharton School at the University of Pennsylvania, explores that very question. The study focused on generating a new consumer product (priced under $50) aimed at college students. To this end, they derived pools of ideas from 35 different prompting strategies and compared these to a baseline set created from human-generated ideas within a Wharton MBA class. Here are the key findings: ➡️ AI-generated ideas tend to lack the "dispersion" that's typically seen in human brainstorming sessions. Dispersion is an assessment of how different or varied ideas are from each other. A higher dispersion implies that the ideas are widely varied and cover a diverse span of concepts or themes. In contrast, low dispersion indicates that ideas are more similar, with less novelty between them. So humans have the edge just now on diversity of ideas (though that doesn't necessarily translate into quality...) ➡️ Prompt Engineering can significantly impact the quality of output from an LLM - as shown in the report by creating more diverse idea sets when carefully engineered prompts are applied. ➡️ Chain-of-Thought (CoT) Prompting: Among the prompting methods evaluated, CoT prompting most effectively increased the diversity of AI-generated ideas, bringing it almost to par with human group brainstorming. CoT not only increased the variance but also produced the highest number of unique ideas. What are the implications of these findings? ➡️ AI idea generation currently falls short of the diversity seen in human brainstorming, but with the right prompting techniques, this limitation can be minimised. ➡️ CoT prompting is particularly effective for generating a wide array of distinct ideas. ➡️ The overlap of ideas between different prompts is small, suggesting that hybrid prompting strategies — in which ideas from multiple prompts are combined — might be an effective approach. According to the report's authors, these results have practical implications for the use of AI in augmenting human creativity, especially in generating more diverse ideas through improved prompting tactics. They conclude that while humans remain slightly ahead, AI, when prompted correctly, it can produce an impressively diverse range of ideas. What do you think? Do you use AI to help generate ideas or support your creative thinking? What prompting techniques do you use to improve the diversity and quality of your ideas? Let me know in the comments below. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpr

What do you get for your £19 to get access to Microsoft CoPilot (the personal version, not including custom CoPilots). Here are my immediate findings. For clarity, this is a very cursory initial test. Then again, anyone encountering Microsoft CoPilot for the first time likely won't have had any handholding or guidance either. As I've done, they'll begin by just experimenting with it. ➡️ Powerpoint: I asked it to generate a Powerpoint presentation based on supplying a basic agenda framework with suggested headings. It returned with a complete slide deck - with slide headings, body content, images, and designed slides. Quite impressive based on a single prompt - with some modest tweaking, this deck could easily be used in anger - so you can immediately see how a lot of time can be saved in generating day-to-day slide presentations. (Is there still a future for all those AI presentation start-ups, e.g. Tome, Beautiful AI, etc). ➡️ Word: Not bad. Good at generating text from a prompt. But not so good at being able to highlight a paragraph and asking it to just work on that alone (or maybe I need to spend more time working out how to get it to do things like this). Having used the GPT for Work extension for Google Docs for nearly a year (and the GPT for Work plug-in for MS Word), perhaps I have higher expectations. GPT for Work is still a nifty bit of kit. And you pay for usage, not users. Not least, being able to pick your LLM of choice, add system prompts, etc. Of course, it's really a personal productivity extension - not the sort of thing you could roll out or manage at scale. ➡️ Excel: Have to say - as per Professor Ethan Mollick - I'm a bit disappointed with CoPilot on Excel. I naively assumed it'd bring Chat GPT Advanced Data Analyis capability to Excel. It most definitely doesn't. Maybe that'll change over time - but right now, my sense is you are going to have to maintain access to ChatGPT+ if you want proper data analysis functionality. I haven't tested Outlook as I'm on a Mac using Apple Mail (and currently no time to set up Outlook for Mac to test out CoPilot). ➡️ Other observations: Microsoft CoPilot response is often very slooooow - and sometimes just falls over (Something went wrong. Please try again in a moment). GPT for Work by contrast is very fast and allows a lot more flexibility in what you can do. If people's first experience of CoPilot is a bit sub-standard - it isn't going to encourage them to keep using it. Also, if users receive little or no guidance on how to use it or what to use it for, then those efficiency and effectiveness gains may be some way off. Has anybody else taken the plunge? What were your first thoughts? Let me know in the comments below. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpr

Are you falling prey to AI's "Asleep at the Wheel" effect? The term was coined by Professor Ethan Mollick on the back of a study he helped conduct with Boston Consulting Group last summer, which appeared to find that consultants over-relying on AI to do their work saw the quality of their output decline. Now, a new report - called "Falling Asleep at the Wheel: Human/AI Collaboration in a Field Experiment on HR Recruiters" - from Fabrizio Dell'Acqua of the Laboratory for Innovation Science at Harvard Business School, sheds more light on the phenomenon. "Falling Asleep at the Wheel" is an investigation into the optimal design of human/AI collaboration in the workplace, with a specific focus on making sure that human workers remain attentive and engaged when working alongside artificial intelligence (AI). Fabrizio Dell'Acqua developed a theoretical model emphasising the trade-off between AI quality and the human effort exerted during tasks. The findings suggest that high-quality AI, while advantageous in isolation, might discourage human effort, leading to a decrease in combined human/AI performance. To test the theoretical model, the author conducted a field experiment involving 181 professional HR recruiters to review 44 resumes each for a software engineering position and make interview recommendations. Recruiters were divided into four groups: those receiving assistance from a perfect AI, a high-performing AI, a lower-performing AI, and no AI assistance. Recruiters were made aware of the treatment they received, and their task accuracy was then measured against objective performance metrics. Here's what the study found: ➡️ Recruiters paired with higher-quality AI were found to be less accurate in their assessments of job applications compared to those paired with lower-quality AI. ➡️ Specifically, those with lower-quality AI exerted more effort and spent more time evaluating resumes, and were unlikely to blindly follow AI recommendations. This group's accuracy was enhanced due to their increased effort. ➡️ Experienced recruiters demonstrated significant improvements in performance when paired with lower-quality AI. In contrast, they underperformed inexperienced recruiters when provided with higher-quality AI, suggesting a complex relationship between human skillset and AI assistance. Dell'Acqua concludes that to maximise human/AI performance, "Bad AI" might sometimes be preferable, making sure human workers aren't over-reliant on AI and remain actively engaged. What do you think? Have you noticed any overreliance on AI? Do you agree that people should be given less powerful AI to work with to mitigate against "asleep at the wheel" effects? Let me know in the comments below. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai #aiinpr

What is the best way to implement AI in a team-based environment? The last week has seen some new options emerge - but what'll work best for you? Until now, the vast majority of AI tools available to PR practitioners have essentially been personal productivity tools. If you had a team of say, 10 people, you could let everyone have a ChatGPT+ account - but effectively you have 10 people all using 10 separate instances of ChatGPT. You could, of course, have created an OpenAI account - but OpenAI's Playground is really a development environment. And doesn't give you the full capabilities of ChatGPT+. Last week, OpenAI launched ChatGPT Team. As its name suggests, this is intended to help organisations use ChatGPT in a more collaborative manner - for a price. Then on Monday, Microsoft made CoPilot available to all organisations who want it (removing the previous requirement to commit to a minimum of 300 licences). All businesses, for $30 per user per month, now get AI embedded into all existing Microsoft office tools, e.g. Word, Excel, Powerpoint, Outlook, etc. You also get the ability to create custom CoPilots (ie custom GPTs). While all of these new options are welcome, it does present somewhat of a dilemma. Patently, most people would prefer to keep the number of vendors to a minimum (as well as keep costs under control). The fact is, no single solution can do everything. For example - if you decided to plump for MS CoPilot and dump ChatGPT+ you won't have access to the latter's Data Analysis capability (early reviews of MS Excel's AI CoPilot suggests it doesn't have the former's AI analytical power). Also, neither appears to address some collaborative features that I think everyone would want eg ability to use multiple document repositories as knowledge augmentors. Or you could stick with ChatGPT+ - but lose the integration with MS tools. Or have both - but then you are paying twice over. Are there alternatives? Typing Mind Custom offers a possible third way. https://lnkd.in/eKArGYdN First - it allows you to use different LLMs - GPT or Claude, with others to follow. So doesn't lock you in to a single LLM. Because it is effectively allowing use via API, you pay based on token usage not users i.e. you'll pay $30 per person for MS CoPilot irrespective of how much or little they use it. Paying by use clearly has benefits - not least in terms of tracking and managing costs. It genuinely allows you to create a shareable prompt library. It also allows you to use documents stored elsewhere as "knowledge" It also offers the option of a self-hosted version, i.e. everything would sit on your own servers. This is one example. There are other options, e.g. Team-GPT. What do you think? How are you hoping to use AI in a team-based context? Let me know in the comments below. 🔔 Visit my profile and "ring" the bell icon for notifications on my latest posts (assuming you find them useful or interesting) #publicrelations #pr #ai