Almost Timely News: 🗞️ 4 Generative AI Trends for the Future
(2024-12-01)
Almost Timely News: 🗞️ 4 Generative AI Trends for the Future (2024-12-01) :: View in Browser
The Big Plug
✅ Download my newly revised Unofficial Guide to the LinkedIn Algorithm, just updated with new information from LinkedIn Engineering!
Content Authenticity Statement
95% of this week's newsletter was generated by me, the human. You'll see an output from Google Gemini in the opening section. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.
Watch This Newsletter On YouTube 📺
Click here for the video 📺 version of this newsletter on YouTube »
Click here for an MP3 audio 🎧 only version »
What's On My Mind: 4 Generative AI Trends For The Future
This week, let's start digging into the 2025 trends in AI. There will be innumerable reports of trends in AI, but we're going to cover probably some of the less obvious or slightly more technical stuff. Let's start with the less obvious.
Part 1: The Rest of the Car
For a while, I've compared generative AI - especially large language models - to the engine of a car. They're incredibly powerful and versatile, capable of nearly any language tasks. I don't need to explain why at this point; almost everyone has used a tool like ChatGPT by now.
What should be obvious, even from just a few minutes' usage, is that the engine of a car is not the same as the car. No one rides down the road on an engine; everyone drives down the road in a vehicle of some kind. And that's where generative AI is today - we're running into the limitations of using an engine directly (through apps like ChatGPT) and needing the rest of the car.
Companies like OpenAI and Anthropic have clearly recognized this need, adding more and more features to their consumer AI interfaces. OpenAI's Canvas, Claude's Computer Use capabilities, and the rise of AI agents all speak to the fundamental reality that most AI use cases call for more than just an engine.
For example, suppose you wanted to know what newsletters you should subscribe to. You could write a detailed prompt about who you are and what you're interested in, then have generative AI build a scoring rubric to assess whether any given newsletter was a good fit for your interests.
That's a great language-based use of generative AI, but fundamentally it still requires a lot of infrastructure that's not present in AI tools. You need a way to connect to your email inbox, to download data from it, and then do something with the results from your AI. All that is the rest of the car.
To the extent that companies want to extract value from AI, they will need the rest of the car. In the months and years ahead, expect more development around the rest of the car to unlock the value of AI.
We already see this with toolkits like LlamaStack, which creates not only an environment to run Meta's Llama model family, but much of the extra core infrastructure needed to use agents, tools, etc.
Expect consumer interfaces to keep adding functionality as well; it would not surprise me in the least to see major AI providers add Zapier-like functionality to their interfaces to speed up adoption of AI. Right now, the major providers have been all-in on making bigger, better, faster models - but there's only so much you can do with a bigger engine without the rest of the car.
What does this mean for you and your business? It's largely good news. When the focus is on building the rest of the car, it means the focus is on making AI more useful and accessible. It's less about the shiny object and more what you can do with AI to make it practical. And the legion of tech vendors that are all competing for more and more scarce dollars can invest time in polishing their AI offerings to make them more useful.
Part 2: Local Models
There are so many different ways to describe local AI models - open models, local AI, etc. They all boil down to the same thing - AI engines you run on hardware or environments under your control.
While most of the headlines in AI have been about the big cloud providers and their foundation models - Google Gemini, OpenAI GPT-4o, Anthropic Claude Sonnet 3.5 - there's been a fast, furious revolution in local AI models as well.
Let's take a look at the leaderboards of many AI benchmark tests as things stand right now:
We see on many different tests that the big foundation models, for the most part, are leading the pack versus local AI models. (any model above with a B in the number is typically a local AI model, like Qwen 2.4 72B) The big tech models are state of the art right now, skilled and capable at many tasks.
What's fascinating, however, is if we wind the clock back a year for the big foundation models. What did things look like roughly this time last year for foundation models, compared to today's local AI models?
What we see is frankly stunning. Today's local AI models - ones you run in your own server room or on a fully-loaded high end laptop - are on par or even exceeding benchmarks from the state of the art models.
That's absolutely stunning when you think about it. Local AI models run on your computers. They require no cloud, no API calls, no third parties - just hosting software you run and the model itself. And yet, in just a year's time, these local models (which are often much smaller than the models provided by cloud providers) have surpassed the state of the art a year ago.
To get the same performance of local models today on a beefy laptop, a year ago you needed a huge server room. That's nuts. To put this in context, imagine buying a luxury car from a dealer last year. Today, the do it yourself kit car you can buy exceeds last year's luxury car in performance. That's how fast AI has evolved in a year's time - and that means that if the big cloud providers ever went out of business, you would still have access to near-state of the art performance and technology in perpetuity. It might not ever get better, but you wouldn't lose access to it.
What are some of the practical use cases of local AI models? As I talked about at MAICON this year, local AI models are as private as the rest of your infrastructure. If you work with sensitive data, using a local model on hardware under your control is the way to go.
But even without sensitive data, there are plenty of use cases for local AI models, especially on the cost side of things. For example, I have a MacBook Pro M1 Max from 2021. It's a solid computer, the best in class at the time. It's capable of running models like Mistral Small or most of Meta's Llama family. I use local models all the time when I'm building a proof of concept that might or might not be commercially viable; by using a local model for development and innovation, the only costs I incur are electricity costs.
Right now I'm working on doing more with Katie's Ideal Customer Profiles, using a local AI model to generate thousands of these things. I've got an idea that I'm working on which needs this level of scale, but I'm unwilling to pay money to a cloud provider for this idea that might never pan out.
The key takeaway here is that everyone should experiment with running local models. Great, free tools like Ollama, AnythingLLM, or LM Studio make this relatively painless, and it's great for both R&D and insurance against a big cloud provider going out of business.
Part 3: The AI Slowdown?
There's been a fair amount of digital ink spilled recently about whether AI model development is slowing down. These sorts of articles always make me scratch my head; AI continues to move at an insane pace of development. Let's take a quick retrospective tour.
Look at JUST Google Gemini.
We started the year with Gemini 1.0 which was basically a relabeled Bard. It sucked.
1.0 became 1.0 Pro in February. Then 1.0 Pro 002 in April. Pro was dramatically better than 1.0.
A month later, 1.5 Pro 001 came out. It was dramatically better than any version of 1.0, including 1.0 Ultra.
Gemini 1.5 Flash 001 also came out in May and it was okay, but kind of dumb.
Four months later, 1.5 Pro 002 came out, and it was dramatically better than 1.5 Pro 001.
At the same time, Gemini 1.5 Flash 002 came out - and it beat Gemini Pro 1.5 001's benchmarks. The CHEAP model in September was better than the Pro model in April.
How about OpenAI?
We started the year with GPT-4, a really solid model.
At the end of January, we got GPT-4-Turbo. Solid, and was the underpinning of many custom GPTs.
In April we got GPT-4-Turbo with vision, able to see visual inputs.
Then in May Turbo got kicked to the curb by GPT-4o, the first natively multimodal model from OpenAI.
4o got two more versions, including last week's big upgrade.
4o-mini came out in July as GPT-3.5-Turbo's replacement.
And in September, o1 preview came out, OpenAI's first model with metacognition.
And Anthropic had a big year, too.
We started the year with Claude 2.1, which was okay but functionally pretty dumb and a pain to work with
Claude 3 came out in March and was a big performance increase, especially Claude 3 Opus.
Claude 3.5 Sonnet came out in June, and is state of the art on many tasks like coding; it got a boost in October with a version 002 that gives it agentic capabilities, like being able to take over your computer with your permission. Claude 3.5 Haiku came out just a couple weeks ago.
What else happened JUST this year?
In April, Meta released Llama 3, which was an incredible open model that you could run on beefy consumer hardware or corporate servers, entirely self hosted. It was the champ of open models for a hot minute, briefly supplanting Mistral Small.
And then Meta dropped the bomb on everyone. In July, Llama 3.1 was released with a state of the art 405 billion parameter model that had equivalent performance to all the closed models. For the first time, companies wanting completely private LLMs could have a state of the art one.
Keeping up with the changes in AI this year was practically a profession unto itself.
Just because there isn't a new model announcement every month from the big providers doesn't mean AI is slowing down. It's proceeding at its usual crazy rate.
What's the key takeaway here for us as individuals and businesses? It means our marketing needs an abstraction layer between us and the AI nuts and bolts. The market is moving so fast, too fast for any vendor to keep up with, absent having an abstraction layer.
Any AI project you work on, any vendor you deal with, needs to have a buffer between the engine and the rest of the car. The engine is changing so fast that if you build a car around a specific version of the engine, that car will be obsolete in months.
Part 4: The Future of Content
Let's take a step back here for this final part, something that requires a bit of reflection. Until the advent of generative AI, if you wanted a particular kind of content, you could generate it, but it took substantial effort. You had to hire people or do it yourself, and the time it took to generate content was measured in hours, days, weeks, or even months.
Generative AI changes that equation substantially. A generative model has the ability, as you well know, to create nearly anything in most common formats. They can write stories, produce music, create images, paint portraits - the use cases are endless, and the model capabilities today are greater than ever, with no sign of slowing down.
Which means that a future where content is effectively on demand isn't far away. Here's what I mean: any idea you have, you can have content for in minutes, if not seconds, and the content will often be good enough.
Suppose you wanted to listen to a particular kind of song, like symphonic rock, but with a fiddle as the focus instrument. A generative model like Suno v4 can create that content for you with reasonable quality. It's not as good as a human composer, but it's exceptionally quick.
I had this experience over the Thanksgiving holiday, when I was driving. I wondered what a symphonic rock piece might sound like with a fiddle, and so when I pulled into a rest area (never AI and drive), I prompted Suno to create that. By the time I was finished at the rest area a few minutes later, I had a playlist of 10 variations to listen to. Two were terrific, four were okay, and 4 sucked. That's pretty incredible - to think that where content didn't previously exist, I as a tech-savvy consumer could have tailored content on demand in just minutes.
I want you to marinate on that sentence for a moment.
As a tech-savvy consumer, you could have tailored content on demand in just minutes.
As marketers and business folks, we have spent decades focused on creating more and more content (of varying quality) in an effort to feed the various content-powered engines of the world, from search engines to social media.
Each time a new content format has come out, it has diluted the market share of other forms of content. Podcasts take some share of ear, as Edison Research calls it, from traditional radio. Spotify took some share of ear from radio. Everything new nibbles away at what came previously. Now, we're not talking about the "death of" some content format, but an acknowledgement that the more choice a consumer has, the harder existing formats and channels have to work to retain their audience.
A future in which the consumer creates the content they want on demand deeply challenges the media landscape. What if I don't have to consume existing content at all? What if generated content is good enough, especially if it's what I ask it for?
Here's another silly example. My friend Ashley shared on her Instagram stories a meme Reel that more or less said, "Imagine a Hallmark movie but it's about a woman who learns the true meaning of Christmas while pursuing a serial killer". When I saw that, I chuckled.
And then on a whim, I opened Google Gemini and with a storywriting prompt, and had the model see if it could generate that kind of story. Here's a portion of what it spit out:
Title: Eagle Valley Angel
Logline: A cynical NYPD detective visiting her family for Christmas in a small Indiana town uncovers a serial killer targeting abusive men, and finds herself falling for the avenging angel she’s sworn to bring to justice.
Part 1: Silent Night, Deadly Night
Scene 1: Big City Blues:
Introduce Sarah, a 38-year-old bisexual NYPD detective, feeling burnt out and cynical in the bustling city. Christmas is approaching, but she feels disconnected from the holiday spirit.
Her boss forces her to take mandatory holiday leave, sending her back to her quaint hometown of Eagle Valley, Indiana. Sarah is reluctant but complies.
Scene 2: Home for the Holidays (with a Twist):
Sarah arrives in Eagle Valley, a picture-perfect Hallmark town decked out in Christmas decorations. Her family (warm, quirky, and typically Hallmark-esque) welcomes her, but Sarah feels out of place.
Introduce key family members: maybe a well-meaning but nosy mom, a supportive but clueless dad, and a younger sibling who embodies the Christmas spirit.
Over breakfast, Sarah reads a local newspaper article about a series of mysterious deaths – all men with reputations for harassing or assaulting women. The local police are baffled, labeling it the work of a possible serial killer.
Scene 3: Small Town, Big Suspicions:
Sarah visits the Eagle Valley Police Department, ostensibly to offer her expertise, but mostly out of boredom and professional curiosity. She finds the local police force charmingly incompetent and ill-equipped to handle a serial killer.
Sarah starts her own unofficial investigation, using her big-city skills to analyze the crime scenes and victim profiles. She notices a pattern: the victims were all confronted shortly before their deaths, and a specific tool (maybe a wrench or tire iron) was used in each case.
Montage: Sarah investigates, visiting crime scenes, interviewing witnesses (town gossips, local business owners), and piecing together clues. The Christmas setting is prominent, creating a contrast between the festive atmosphere and the dark nature of the crimes.
Scene 4: The Angel Appears (From Afar):
Introduce Alice indirectly. Sarah finds a piece of evidence at one of the crime scenes – a small, handmade angel ornament, perhaps dropped accidentally. This becomes Alice’s “signature.”
Sarah researches the angel ornament, tracing it back to a local craft fair. She starts to suspect the killer is a woman, and one who is skilled with tools.
That's only part of the story, but the reality here is that this took less than 10 minutes to produce. It's content on demand. Gemini was able to write the full draft of the story in about 30 minutes, scene by scene, and it was good enough. It was the equivalent of a Hallmark movie - not high art, not going to win a Nobel Prize in literature by any stretch, but just as good as what you'd watch on TV or the streaming service of your choice.
That's content on demand - content exactly the way a consumer wants it, when they want it, limited only by their ability to verbalize what it is they want.
Here's the key takeaway to think about as a marketer: in that landscape of infinite choice and getting exactly what you want, why would a consumer ever want to consume our marketing content? If you have a choice between someone else's marketing content and content that exactly and precisely satisfies your needs, why would you ever choose the marketing content unless you had a specific need?
The solution for us marketers is straightforward: before this future arrives, we absolutely must build audiences for ourselves and provide them what they want, what they need, what entertains and educates them. If we don't do this, if we don't build loyal audiences today, we might find ourselves with no audience at all tomorrow.
We already see hints of this with AI-powered search, where consumers can talk to ChatGPT or Perplexity or Google Gemini and get synthesized answers (which send almost no traffic to us poor marketers), but that's just the tip of the iceberg. A future of generated content on demand that exactly meets the needs of the consumer crowds out everyone else in the media landscape that doesn't exactly meet the needs of the consumer.
Wrapping Up
For now, this is a start to some of the trends in AI that we'll be encountering in the near future. I'll have some more to share in the next couple of issues, such as on our digital ghosts, AI agents, and more, but this is a good starting point.
Take each of the key takeaways and think them through. How will they apply to your work, your organization, your future? Take the necessary steps you need to build proficiency for each takeaway, and you'll stay ahead of the curve.
(and if you'd like content like this delivered as a workshop at your company or event, let me know.)
How Was This Issue?
Rate this week's newsletter issue with a single click. Your feedback over time helps me figure out what content to create for you.
Here's The Unsubscribe
It took me a while to find a convenient way to link it up, but here's how to get to the unsubscribe.
If you don't see anything, here's the text link to copy and paste:
https://almosttimely.substack.com/action/disable_email
Share With a Friend or Colleague
If you enjoy this newsletter and want to share it with a friend/colleague, please do. Send this URL to your friend/colleague:
https://www.christopherspenn.com/newsletter
For enrolled subscribers on Substack, there are referral rewards if you refer 100, 200, or 300 other readers. Visit the Leaderboard here.
Advertisement: Bring Me In To Speak At Your Event
Elevate your next conference or corporate retreat with a customized keynote on the practical applications of AI. I deliver fresh insights tailored to your audience's industry and challenges, equipping your attendees with actionable resources and real-world knowledge to navigate the evolving AI landscape.
👉 If this sounds good to you, click/tap here to grab 15 minutes with the team to talk over your event's specific needs.
If you'd like to see more, here are:
ICYMI: In Case You Missed it
Besides the Generative AI for Marketers course I'm relentlessly flogging, this week, there was almost nothing because of the holiday!
Almost Timely News: 🗞️ 4 Reasons Why Generative AI Prompts Fail (2024-12-01)
Mind Readings: Why Generative AI Content is Bland and How To Fix It
You Ask, I Answer: Ethics of Using Other People’s Data in Generative AI?
In-Ear Insights: Marketing Lessons From MarketingProfs B2B Forum
So What? Q4 2024 Large Language Model Comparison and Bakeoff
Skill Up With Classes
These are just a few of the classes I have available over at the Trust Insights website that you can take.
Premium
Free
Generative AI for Architecture, Engineering, and Construction
Powering Up Your LinkedIn Profile (For Job Hunters) 2023 Edition
Predictive Analytics and Generative AI for Travel, Tourism, and Hospitality, 2024 Edition
Building the Data-Driven, AI-Powered Customer Journey for Retail and Ecommerce, 2024 Edition
The Marketing Singularity: How Generative AI Means the End of Marketing As We Knew It
Advertisement: Generative AI Workshops & Courses
Imagine a world where your marketing strategies are supercharged by the most cutting-edge technology available – Generative AI. Generative AI has the potential to save you incredible amounts of time and money, and you have the opportunity to be at the forefront. Get up to speed on using generative AI in your business in a thoughtful way with Trust Insights' new offering, Generative AI for Marketers, which comes in two flavors, workshops and a course.
Workshops: Offer the Generative AI for Marketers half and full day workshops at your company. These hands-on sessions are packed with exercises, resources and practical tips that you can implement immediately.
👉 Click/tap here to book a workshop
Course: We’ve turned our most popular full-day workshop into a self-paced course. Use discount code ALMOSTTIMELY for $50 off the course tuition.
👉 Click/tap here to pre-register for the course
If you work at a company or organization that wants to do bulk licensing, let me know!
Get Back to Work
Folks who post jobs in the free Analytics for Marketers Slack community may have those jobs shared here, too. If you're looking for work, check out these recent open positions, and check out the Slack group for the comprehensive list.
Advertisement: Free Generative AI Cheat Sheets
Grab the Trust Insights cheat sheet bundle with the RACE Prompt Engineering framework, the PARE prompt refinement framework, and the TRIPS AI task identification framework AND worksheet, all in one convenient bundle, the generative AI power pack!
Download the bundle now for free!
How to Stay in Touch
Let's make sure we're connected in the places it suits you best. Here's where you can find different content:
My blog - daily videos, blog posts, and podcast episodes
My YouTube channel - daily videos, conference talks, and all things video
My company, Trust Insights - marketing analytics help
My podcast, Marketing over Coffee - weekly episodes of what's worth noting in marketing
My second podcast, In-Ear Insights - the Trust Insights weekly podcast focused on data and analytics
On Threads - random personal stuff and chaos
On LinkedIn - daily videos and news
On Instagram - personal photos and travels
My free Slack discussion forum, Analytics for Marketers - open conversations about marketing and analytics
Listen to my theme song as a new single:
Advertisement: Ukraine 🇺🇦 Humanitarian Fund
The war to free Ukraine continues. If you'd like to support humanitarian efforts in Ukraine, the Ukrainian government has set up a special portal, United24, to help make contributing easy. The effort to free Ukraine from Russia's illegal invasion needs your ongoing support.
👉 Donate today to the Ukraine Humanitarian Relief Fund »
Events I'll Be At
Here are the public events where I'm speaking and attending. Say hi if you're at an event also:
Social Media Marketing World, San Diego, April 2025
Content Jam, Chicago, April 2025
SMPS, Columbus, August 2025
There are also private events that aren't open to the public.
If you're an event organizer, let me help your event shine. Visit my speaking page for more details.
Can't be at an event? Stop by my private Slack group instead, Analytics for Marketers.
Required Disclosures
Events with links have purchased sponsorships in this newsletter and as a result, I receive direct financial compensation for promoting them.
Advertisements in this newsletter have paid to be promoted, and as a result, I receive direct financial compensation for promoting them.
My company, Trust Insights, maintains business partnerships with companies including, but not limited to, IBM, Cisco Systems, Amazon, Talkwalker, MarketingProfs, MarketMuse, Agorapulse, Hubspot, Informa, Demandbase, The Marketing AI Institute, and others. While links shared from partners are not explicit endorsements, nor do they directly financially benefit Trust Insights, a commercial relationship exists for which Trust Insights may receive indirect financial benefit, and thus I may receive indirect financial benefit from them as well.
Thank You
Thanks for subscribing and reading this far. I appreciate it. As always, thank you for your support, your attention, and your kindness.
See you next week,
Christopher S. Penn