The Generative Economy

It’s transforming before our eyes and we haven’t pre-trained for this. So, I’m going to try and give you a sense of what’s happening just over the last week or so… but there is a lot to get through so I’ll try to keep it brief with my reflections.

???The MerrAI-go-round

Even if you didn’t have a chance to flick through the State of AI 2024 report I shared last week, you’ll know that the amount of investment pouring into AI right now is eye watering:

This seems to be making the industry quite turbulent, and as a result, there is a lot of jostling. Not least at OpenAI:

There seems to be differing opinions on the why of it all, personally I think we’re still in the early stages and there is so much opportunity; now is not the time to stand still, it’s the time to chase dreams!

It’s old news that Ilya Sutskever raised $1Bn to create his new startup and now Mira Murati is looking to raise $100M for her new venture.

Not long ago DeepMind cofounder Mustafa Suleyman joined Microsoft, leaving Inflection, the startup he founded, behind. Now Sebastien Bubeck, who you might know from the paper “Textbooks are all you need” and “Sparks of AGI”, is leaving Microsoft for OpenAI to “further his work toward developing AGI”

For a final spin of this section, Google are reorganising (again) to streamline AI initiatives, with the Gemini app team joining DeepMind. This is an interesting move as DeepMind are ostensibly the research team, albeit a huge one, so moving product there is something to keep an eye on.

????Compute

Nvidia’s stock continues to trend towards the moon because it has deals with all the Hyperscalers and everyone wants to build new data centres (more on that in a moment).

Far from stopping there, Nvidia casually dropped a GPT-4 beating LLM called Llama-3.1-Nemotron-70B-Instruct by using (as the name would suggest) Meta’s open source Llama 3.1.

???Top tip - if you want to play with this new model, check it out on HuggingChat by selecting the settings button in the “Current Model” box:

???Power

We’re gonna need a lot of power for all that compute and new data centres.

The market says, go long uranium (not investment advice!)

“Small Modular Reactors” have been a topic of discussion for some time, many are currently being rolled out in China and India. Of course nuclear has been a challenging topic in the West with the story mostly being around shutting down reactors.

Despite this, “Google has signed a “world first” deal to buy energy from a fleet of mini nuclear reactors to generate the power needed for the rise in use of artificial intelligence.”

Microsoft have “*signed a power deal to help resurrect a unit of the Three Mile Island nuclear plant in Pennsylvania”.*

And Amazon “has signed three agreements to support the development of Small Modular Reactors (SMRs)”

Realistically, meeting carbon emissions goals and staying competitive from an economic perspective is very likely to drive the incentive for more nuclear energy.

???Search

Google’s search business is under attack as behaviour patterns shift and advertising spending adapts. The Innovator’s dilemma is real.

One of the companies taking advantage is Perplexity, with a fresh approach. Last week they added new Internal Knowledge Search capabilities and Spaces for AI-powered research and collaboration.

What caught my eye was CEO Aravind Srinivas’ tweet outlining the categories that Perplexity lacks vs Google, but they will go after:

It’s not just Perplexity though! Stanford University’s STORM is a LLM system that writes Wikipedia-like articles from scratch based on Internet search. Let me show you what it does:

Here is a link to the documentation on GitHub: https://github.com/stanford-oval/storm

???♂??Agents

Since I mentioned them at the end of the STORM demo video I thought I’d follow on with one of the hottest areas in AI at the moment.

Langchain published an interesting post on Memory for Agents

Langchain are also more broadly focused on multi agent support in LangGraph

OpenAI have have released Swarm (experimental, educational) for Multi-Agent Orchestration

With all the talk of agents, proof of humanity potentially become a pressing issue. Many this is why Sam Altman’s Worldcoin felt it was a good time to rename to “World”. It seems this was mostly about the new iris scanning Orb.

I’ll be honest, this gives Portal vibes which, whilst a great game, I am not sure even I like the idea of IRL.

???Tools

Adobe MAX happened last week and weirdly seems to have been relatively under the radar despite the great demos. For me they have absolutely nailed the AI value proposition.

In the demo below they’ve added a simple one-click tool which would likely have taken an artist several hours to redraw into the various combinations of 2D vectors. To boot, they have likely unlocked a huge amount of potential new creativity that would not have previously been possible.

Check out all of the Adobe MAX Sneaks at this playlist.

You all know I’m a NotebookLM fan. The team have recently enabled customisation of their viral Audio Overviews feature and they have also just published a great guide on using the tool. But here is a recent presentation by Editorial Director at Google Labs, Steven Johnson, on AI + Education 2.0:

Finally, HeyGen launched Interactive Avatar with a variety of template avatars and use cases. Even been double booked and wanted to be at both meetings? Or struggled to find convenient times to meet across timezones?

Maybe AI-you can attend… I am sure this might raise some A-eyebrows.

???Regulation

Maybe someone sent Emmanuel Macron a link to my newsletter ??

And finally, an interesting article in Forbes highlighting some of the challenges at an industry level: Why A Lawsuit In Massachusetts Is A Wake-Up Call For Education

Phew.

As always, thanks for reading.

Jody