Almost Timely News: The Next Wave of Generative AI (2023-05-07) :: View in Browser

Almost Timely News

👉 Watch my brand new keynote, The Marketing Singularity, all about how generative AI is the end of marketing as we know it »

Content Authenticity Statement

97% of this newsletter was written by me, the human. There are two screenshots of AI-generated content.

Watch This Newsletter On YouTube 📺

Almost Timely News: The Next Wave of Generative AI (2023-05-07)

Click here for the video 📺 version of this newsletter on YouTube »

Click here for an MP3 audio 🎧 only version »

What’s On My Mind: The Next Wave of Generative AI

This week, let’s talk about what’s happening right now in generative AI, because it’s been a big week. Well, it’s been a big few weeks, so let’s go over what those developments mean. As you know, last fall, OpenAI released its language model interface, ChatGPT, that opened the door for non-technical users to be productive with large language models. The model – and remember in the context of AI, a model is just a fancy word for software – behind ChatGPT is a massive behemoth known originally as InstructGPT.

These models are large, very expensive to train, and costly to operate. For years, other developers and companies have tried making their own, but the costs of starting from scratch, assembling the massive quantities of data needed to train (build) a model, and deploying it are usually well out of reach of scrappy entrepreneurs. There have been many attempts and starts over the years but none have been able to perform as well as the big money models that big tech companies created. Thus, for many companies and many people like you and me, ChatGPT has been the only serious game in town.

Until about a month ago. Facebook/Meta released their own model, LLaMa, but in a different way than the other tech companies. Rather than give away an interface like Bing or Bard or ChatGPT, they released the underlying model, LLaMa, itself as non-commercial open source software. LLaMa is the same high quality as the other big tech models, but it’s available to many more people for free. This is a big deal because Facebook basically took this gigantic model trained on a trillion words and just… gave it away.

That was the first pebble in the avalanche.

In AI, there’s a concept called fine-tuning, where you take an existing model and tailor it to your needs. Remember that these language models don’t contain actual words. They contain mathematical probabilities about words, like a giant library of statistics about what words are near other words, what phrases are near other phrases, etc. A big public general model like the ones from OpenAI are gargantuan because they have to be a jack of all trades, kind of like the family dog. Part companion, part retriever, part guard dog, and not overly specialized at any one thing. When we want a language model to do one thing very specifically, we change the probabilities in its library to overly favor one thing over anything else. That’s like training a dog to specifically be a bomb sniffing dog; the dog will not be good at sniffing for drugs or earthquake survivors, and probably won’t be as suited for other general dog tasks.

Fine-tuning a model isn’t nearly as costly as building the model in the first place. If entrepreneurs and engineers wanted a custom model for a specific task, it’s far easier to fine tune an existing model, as long as the source model is high enough quality. And that’s what LLaMa is – a very high quality starting point for a lot of innovation that Facebook released to the world. Think of LLaMa like this: let’s pretend that generative AI is like pizza. Up until now, you had to order pizza delivery from OpenAI, right? Through ChatGPT and their APIs, they were the only game in town. You might have thought about making your own pizza from scratch, but for a variety of reasons – time, money, talent – you just didn’t. Along comes Facebook and LLaMa, which is like one of those pre-baked pizza kits. Now all you have to do is customize the very nice pre-made pizza with the toppings you want, but you don’t have to go through all the work of making the pizza from scratch.

In the several weeks since LLaMa came out, we have seen a massive explosion of new derived models, models that are very high performance but scaled to run on hardware as small as a hobbyist’s Raspberry Pi. The tuning capabilities are robust; we see models tuned specifically for tasks like research, healthcare advice, finance, and more. That’s what an open source model enables – massive variation, massive diversity in the space.

There are even projects to put these models on your laptop as private chat instances, like the GPT4ALL software. This looks and runs like ChatGPT, but it’s a desktop app that doesn’t need an internet connection once it’s set up and, critically, it does not share data outside your individual computer, ensuring privacy. Up until now, services like ChatGPT have sent your data to a third party company for use, which is why we’ve said you should never, ever use them with sensitive information. Now, that’s no longer the case – you can use GPT4ALL in complete privacy. It’s the best of both worlds – the performance and capabilities of a service like ChatGPT with ironclad privacy because the data – your data – never leaves your computer. That makes it ideal for industries like finance, healthcare, government – any place where you wouldn’t just want to hand over protected information willy nilly.

Screenshot of GPT4ALL

This has made big waves in the tech community; a post recently by a Google employee has made waves by declaring that neither Google nor OpenAI are paying enough attention to open source, and the open source movement is racing past the big tech players with their closed models. I agree with the engineer’s assessment; open source is a powerful movement that democratizes technology and makes it accessible to almost anyone. There’s a reason Linux – the open source operating system – power a majority of the public internet servers. It’s better, faster, more secure when operated correctly, and near zero cost. The same is now happening in AI.

Why did Facebook do this? Why did they give away such a valuable piece of intellectual property? Because they’re behind. Their most recent efforts in AI have not gone well. So rather than try to do it themselves, they’ve simply done the hard grind of assembling the model and then tossed it to the community, to the world, to do with as we please – and already, coders and developers have taken their excellent base model and made insane improvements in a very short time. There are advancements that take Facebook’s base model and tune it for chat, tune it to be multiple times faster, tune it to run on nearly any device. The community, in effect, did all of Facebook’s R&D for free.

So that’s what’s happening. Let’s talk about what this means, for marketing and for society overall. First, let’s dig into the marketing side. Previously, to deploy a large language model in a marketing context like a chatbot on your website, you pretty much had to pay the OpenAI tax and use their APIs if you wanted high quality output. With the release of LLaMa and the crazy number of free, open source models (including some derivatives that are licensed for commercial use), that’s no longer the case. Now, if you have the technical team in place, you can use an open source model and save yourself a big bucket of money.

Anyone who’s marketing software? Building a large language model into your software just got a whole lot easier and more privacy-compliant, not to mention nearly free. Instead of having to wrestle with commercial licensing and privacy controls, you can now just stuff an open source model into your software and run it locally with no privacy issues. OpenAI API fees? Those just went to zero for software companies. That’s a big win for software companies – especially scrappy startups – and for us consumers who use those products.

For marketers who are just getting used to ChatGPT, this is also a boon. You can have a model that runs on your desktop or laptop computer and has 95% of the performance of ChatGPT with none of the privacy issues – and has a stable underlying model that your company can control. If you’ve ever used ChatGPT after they upgrade the underlying model, you’ve probably noticed that once-reliable prompts get wonky for a little while. This explosion of open source models means you can freeze which model you’re using until you and your organization are ready to upgrade. It’s under your control, which is a big deal.

For marketers who work in regulated industries or secure workplaces that have been forbidden to use ChatGPT, this is now an avenue for you to approach your IT department and explain how this open source movement will let you have the benefits without the risks.

For marketers who have access to technical resources that can fine-tune these open source models, that’s where you’re going to see massive benefit. These models are relatively straightforward to fine-tune. (not easy, but simple) It’s now even easier to customize them to your company, to your needs, to fulfill specific tasks that your team needs to work on. If you recall from the keynote address I’ve given, the more fine-tuned a model is, the shorter and less cumbersome your prompts have to be. You can imagine a set of different task-based models available to you in your job.

And for marketers who are late to the game with large language models, this is unfortunately going to muddy the waters some because each model itself is different – including what prompts do and don’t work with them. Vicuna-13B or LLaMa-30B can operate as powerfully as ChatGPT’s GPT-3.5-Turbo model, but they have a different prompt structure, so you’ll want to pick a platform and learn it before hopping from platform to platform. My recommendation would be for a marketer just getting started to start with ChatGPT for a few months and then move to GPT4ALL with the Snoozy 13B model, as it’s very capable.

Now, let’s talk about the big picture, because it’s worth a discussion. The LLaMa model is incredibly powerful, on par with the GPT models from OpenAI. There are versions that have no restrictions of any kind on them, versions you can ask nearly any kind of question and get a coherent answer, even if that answer is horrifying. Software is inherently amoral. It’s a tool, and thus how that tool is used depends on who’s using the tool. Here’s an example, redacted, about making something you really shouldn’t make at home:

Redacted image of forbidden content

OpenAI will tell you absolutely not, under no circumstances will it answer this question. An unrestricted model gives you an answer (though it’s just as likely to be incorrect as ChatGPT).

There will be misuses of these open source models, just as there are people who use open source website software like Apache to run websites filled with hate and bigotry. These tools will enable content creation of all kinds, good and bad, and we need to be prepared for what that looks like. Here in the USA, next year is a presidential election year and I have absolutely no doubt that hostile parties like Russia will attempt to interfere in our elections (as they have in the past) using tools like these to create massive amounts of disinformation, manipulating easily-persuaded people.

But that would have happened anyway. A hostile nation-state like Russia has the resources to build custom models from scratch. These models just make the process faster for everyone, good and bad alike.

And these models, particularly the unrestricted ones, do enable greater positive uses as well. There’s some content that closed models like ChatGPT flat out will not create, even though that content might have legitimate artistic value, like explicit literature, or controversial writing about sensitive topics. Do people who want to write about those sorts of things have the right to do so? Yes. Can they with the current closed source ecosystems? No. So these models will enable that as well.

What we should expect to see, what we are already seeing, is a massive explosion in the use of large language models. We should expect to see these models showing up everywhere, embedded in software we use all the time – now made free and more accessible. I believe that will overall be a net positive, even though they come with significant downsides you just can’t hand-wave away. Like the Internet itself, like the smartphone, like the personal computer, this new generation of AI models amplifies humanity. What’s good about us becomes better, what’s bad about us becomes worse.

No matter what, the reality is that large language models are now very definitely here to stay. A company like OpenAI could go out of business. Now that open source software exists that is rich, robust, and capable, all the big AI companies could vanish tomorrow but the technology is available in everyone’s hands.

Finally, this also has one other major effect. Open source software is nearly impossible to regulate because in many cases, there’s no central entity in charge of it that has the power to turn it off. The Apache Foundation has zero ability to turn off anyone who’s using their software as a webserver. Mozilla can’t turn off Mozilla browsers around the world. The Linux Foundation has no control over millions of servers and desktops running the Linux OS. That means any legislation, any governmental regulation of large language models will need to focus on the effects, on the outputs, on what people do with the tools because it’s no longer possible to regulate the tools themselves. It’s highly likely legislators and elected officials don’t understand this at all, and they will need to, very soon.

The tidal wave of generative AI has picked up pace. We can either surf it, or drown in it, but either way, there’s no stopping it.

Got a Question? Hit Reply

I do actually read the replies.

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

ICYMI: In Case You Missed it

Besides the newly-refreshed Google Analytics 4 course I’m relentlessly promoting (sorry not sorry), I recommend the livestream from this week where we demoed how to fine-tune a large language model like GPT-3.

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

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: LinkedIn For Job Seekers & Personal Branding

It’s kind of rough out there with new headlines every day announcing tens of thousands of layoffs. To help a little, I put together a new edition of the Trust Insights Power Up Your LinkedIn course, totally for free.

👉 Click/tap here to take the free course at Trust Insights Academy

What makes this course different? Here’s the thing about LinkedIn. Unlike other social networks, LinkedIn’s engineers regularly publish very technical papers about exactly how LinkedIn works. I read the papers, put all the clues together about the different algorithms that make LinkedIn work, and then create advice based on those technical clues. So I’m a lot more confident in suggestions about what works on LinkedIn because of that firsthand information than other social networks.

If you find it valuable, please share it with anyone who might need help tuning up their LinkedIn efforts for things like job hunting.

What I’m Reading: Your Stuff

Let’s look at the most interesting content from around the web on topics you care about, some of which you might have even written.

Social Media Marketing

Media and Content

SEO, Google, and Paid Media

Advertisement: Google Analytics 4 for Marketers (UPDATED)

I heard you loud and clear. On Slack, in surveys, at events, you’ve said you want one thing more than anything else: Google Analytics 4 training. I heard you, and I’ve got you covered. The new Trust Insights Google Analytics 4 For Marketers Course is the comprehensive training solution that will get you up to speed thoroughly in Google Analytics 4.

What makes this different than other training courses?

  • You’ll learn how Google Tag Manager and Google Data Studio form the essential companion pieces to Google Analytics 4, and how to use them all together
  • You’ll learn how marketers specifically should use Google Analytics 4, including the new Explore Hub with real world applications and use cases
  • You’ll learn how to determine if a migration was done correctly, and especially what things are likely to go wrong
  • You’ll even learn how to hire (or be hired) for Google Analytics 4 talent specifically, not just general Google Analytics
  • And finally, you’ll learn how to rearrange Google Analytics 4’s menus to be a lot more sensible because that bothers everyone

With more than 5 hours of content across 17 lessons, plus templates, spreadsheets, transcripts, and certificates of completion, you’ll master Google Analytics 4 in ways no other course can teach you.

If you already signed up for this course in the past, Chapter 8 on Google Analytics 4 configuration was JUST refreshed, so be sure to sign back in and take Chapter 8 again!

👉 Click/tap here to enroll today »

Tools, Machine Learning, and AI

Analytics, Stats, and Data Science

All Things IBM

Dealer’s Choice : Random Stuff

Advertisement: Ukraine 🇺🇦 Humanitarian Fund

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 our ongoing support.

👉 Donate today to the Ukraine Humanitarian Relief Fund »

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:

Events I’ll Be At

Here’s where I’m speaking and attending. Say hi if you’re at an event also:

  • B2B Ignite, Chicago, May 2023
  • MAICON, Cleveland, July 2023
  • ISBM, Chicago, September 2023
  • Content Marketing World, DC, September 2023
  • MarketingProfs B2B Forum, Boston, October 2023

Events marked with a physical location may become virtual if conditions and safety warrant it.

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


You might also enjoy:


Want to read more like this from Christopher Penn? Get updates here:

subscribe to my newsletter here


AI for Marketers Book
Take my Generative AI for Marketers course!

Analytics for Marketers Discussion Group
Join my Analytics for Marketers Slack Group!