Navigating LLMs: Hot Takes on the Most Popular Large Language Models

AI is everywhere now, and the race is on to declare a winner in the AI WARS!

Not really. I’m lying. Large Language Models, the most common form of AI you’re seeing hype about right now is everywhere, and will continue to get baked into practically everything you do online, for better or for worse.

In this post I’d like to explore some of the more popular tools, and their advantages or disadvantages in the ecosystem. Our subscription budget at VIA Studio/eResources has exploded in the last 6 months as we test and try out these tools, so here go some hot takes.

OpenAI: In what’s becoming myth and legend in this world, OpenAI added a “chat” interface to its LLM and immediate value was birthed. They launched early, to some weird results, but is always improving. I use OpenAI almost daily for certain tasks.

Hot takes:

• OpenAI doesn’t have the baggage of the bigger players, so I hope to see them continue to innovate in user interface design.
  
  
• OpenAI is likely to become somewhat of a utility, in that many, many startups will use OpenAI to power their business. Much like the big players have done over the years, they will have the insight to either purchase these startups and incorporate them into their business, or build similar functionality.
  
  
• ChatGPT is TERRIBLE at being creative. Trying to get it to move away from cliches in writing and ideation is a barrier to some of the work that we do. Creating something unique and novel is still a human’s job.
  
  
• Dall-E, OpenAI’s image generator, has caught up and surpassed MidJourney’s image generation in many ways. I absolutely LOVE Dall-E’s ability to take feedback and tweak a specific image. So much so that I used it to help write a fun little fable/comic. You can see it in unfinished format here.

Google Gemini: Google was an early innovator in this space, so much so that they had to fire one of their employees for claiming they created something sentient. Do I think we’ve created Artificial General Intelligence yet? No, I think that there’s just so much human language and sentiment now online that the models have utilized it to blaze past the Turing Test. Now we need new gauges of “intelligence” in these systems.

Hot takes:

• Google has access to your writing style, your email, your calendar, your search history, and a whole bunch more. Watch out, because once they create pipelines to connect these things, Gemini will become scary useful. This is the functionality that will both free up a lot of our time AND cost a lot of jobs in the information economy.
  
  
• Gemini seems to be falling into their larger design patterns, so I worry that they won’t innovate in the user interface space as quickly as OpenAI.
  
  
• Google Search SUCKS. As Google’s biggest cash cow, I wonder how fast they are transitioning to new revenue streams. Google’s SERPs are just a horrible mess these days, with ads everywhere and its usefulness waning. I’m dying to see if they evolve fast enough to keep from becoming a dinosaur.

Microsoft: As much as my design friends hate to hear it, I’m a Microsoft guy when it comes to computers. Ever since Adobe’s products have worked the same on Windows and Mac, it seemed absurd to own a Mac. Sorry, not sorry.

Hot takes:

• Microsoft has already integrated their Copilot functionality to the Windows operating system interface. They have access to millions, if not billions of computers, so their ability to integrate LLM into your workflow tools has high potential.
  
  
• Lest we forget, Microsoft owns LinkedIn, XBox, GitHub, and more, so their ability to get to know you, your career, what’s hot in gaming, and coding models will continue to evolve and improve. I think they will stay near the top of the pack.
  
  
• Microsoft owns a significant stake in OpenAI, and flexed their muscles wildly when the OpenAI board tried to remove Sam Altman. Their influence over OpenAI will keep them as a key player in this space.

Midjourney: Midjourney is freaking awesome at generating all kinds of images and styles. I love it. However, it’s got problems.

Hot Takes

• I think MidJourney, unless it is purchased by a larger corporation, is ripe for class action lawsuits around copyright infringement. I personally believe that we will need new regulations around fair-use and trademark issues around LLM image generators, but it will take lawsuits to get Congress to figure this out.
  
  
• Midjourney’s model of using Discord as their user interface is brilliant. Being able to watch people from all over the world learn their prompts in real-time is both educational and inspiring. I encourage you to watch and learn there.

Adobe Firefly: As a heavy user of Adobe, I also hate their monopoly power over creative tools. Their pricing model sucks as well. Other than that, Firefly is growing up quickly. I still prefer Dall-E and Midjourney, but I’m guessing Firefly will take off. The Firefly web interface is very compelling.

• They have already integrated LLM/AI tools into their products. I’ve used Photoshop and Illustrator’s generative tools quite a bit already. They’re not great, but they can only get better from here. Expect to see more very quickly.

Meeting Notetakers

Currently our favorite AI note taker is MeetGeek, but Google Meet, Zoom, Teams and the rest are already starting to integrate this functionality into their tools. Here’s hoping these evolve quickly and let you tweak them to your needs, because they are already super-useful.

That does it for now. If you have your own hot takes, why don’t you hop over to my LinkedIn post and join the conversation?

Cheers! ~jason

PS - I had Google Gemini review this post and it did NOT like my comment about search dying:

“Accuracy: Some claims, like Google Search dying, are debatable and require nuanced discussion.” (LOL)