Two non-obvious AI milestones I’m tracking

Trying not to stare directly at the sun

Much of AI attention (and investment) is focused on the foundational models and data centres but important contributions to progress and adoption happen elsewhere too (which tend to be more reasonably priced). A couple of recent developments on my radar:

Enablement

On Tuesday, Accenture announced they were acquiring Faculty (no official price but rumours of ~$1b which would be a remarkable multiple). Regular readers will know that I am very bullish on the AI Enablement space (which Max, Daria and I are focusing on with our 10xHumans strategy). I have very high conviction that a) AI will be long-term transformative but also b) that it will require a huge effort in training, consulting and implementation to help business get there. I was quite interested to see Dwarkesh publish an essay in December more or less saying the same thing:

Models keep getting more impressive at the rate the short timelines people predict, but more useful at the rate the long timelines people predict.

I’ve spoken to about a hundred founders of AI enablement consultancies over the last few months. There is a qualitative difference between native AI founder-led enablement firms and those which are pre-LLM. Speaking with enablement clients across Europe, it’s pretty common to hear them describe the main consulting firms (Bain, Accenture, McK) as too old-school: ‘it’s the same people who were giving data-lake strategy presentations now talking about LLMs’.

It was a good move by Accenture to acquire Faculty and potentially a genius one to make Marc their global CTO.

Reversibility

I’ve been a fan of what Rodolfo and Hannah are building at Vaire Computing before it even had a name. You have probably never heard of reversible computing but given the trajectory of the token economy, it’s likely you will be reading about it a lot more (Nvidia acquiring Groq is just the beginning of chip specialisation). Summarised to the point of inaccuracy, reversible computing is basically a very different (and historically mostly ignored) paradigm which Vaire are using to build low-power chips. The company has managed to not only gather most of the world’s reversible computing research talent but also deliver a viable tape-out.

A few weeks back Vaire quietly announced that both Stephen Wolfram (creator of Mathematica, Wolfram Alpha, and the Wolfram Physics Project) and Durgesh Srivastava (previously NVIDIA’s data center product architect) had joined their technology advisory board. I claim no semiconductor expertise outside of reading Dylan Patel’s newsletter but there is clearly something very interesting going on here.