MemeAGI
With GPT 4o’s new image capabilities, it feels we’ve finally reached memeAGI.
This is the world we live in, we ship a ridiculously impressive, almost unbelievable AI innovation, and people use it almost exclusively to anime-ify the world.
But jokes aside, that’s a massive, massive leap for OpenAI’s image capabilities, which were way behind the rest in recent months - and keeps alive the tension between foundation cos and wrapper cos. It seems like each week we swing from one to the other.
When the large model companies push a new model out, everyone is extremely impressed and thinks it’s impossible to compete with them.
Then, within weeks, wrapper companies will build on top of that model and make incredible progress with 1/1000th the capex required, and everyone gets convinced the value is on the application layer.
By the way, not to be outdone, Google released Gemini 2.5, their most intelligent model to date.
Missing revenues, spies, China-made chips
Venture funding is always somewhere around a faithful instantiation of the power law and a hype-chasing game. 31% of VC funding concentrated in 20 mega deals last year, with Databricks, OpenAI, XAI and Anthropic taking the lion's share.
But it's still a risky game. Shortly after announcing AI sales rep Julian, recent industry darling 11x found itself in the barrel of a TechCrunch article revealing it may (or may not) have misrepresented revenues and customers. The company raised a $24m series A from Benchmark and other investors, followed by a $50m Series B led by a16z in September last year. 11x founder Hasan Sukkar denied most of the claims. And investors said it's a straight-up lie.
A few days earlier, another spy story shook the venture world. Human resource platform Rippling's CEO, Parker Conrad, claimed that competitor Deel planted a spy within Rippling to steal trade secrets, and unveiled how they figured it out.
But moving to the real spy world, we have China unveiling deep-sea cable-cutting technology, and a few things to flag on all sides of the Pacific and Atlantic:
Michael Pettis on China's economic challenges and trade war risks
One thread on Sinofuturism, from desert greening to lunar stations
Ant Group claims it's using high-performing China-made AI chips (some details if you don't have the Bloomberg subscription, here). Overall a good week for Alibaba.
And one story we loved reading on Ivory Coast's economic success.
From the physical world
Caleb Watney, co-CEO of the Institute for Progress, wrote a guest essay on what it means to do science. Yes, it is about return on investment. Yes, it is about making funding more efficient so resources are allocated where there's more leverage. But cost-cutting is not everything - "If a venture capital firm measured efficiency purely by how little money it spent, rather than by the returns it generated, it wouldn’t last long." The reality is that there's a curve with a maximum of societal ROI, which at some point decreases as capital allocation gets too haphazard. We must keep thinking of what good science means, and what efficient funding is.
Yet more odd quantum computing news
The New Scientist reports skepticism regarding D-Wave’s quantum supremacy claims. The piece highlights that the lack of rigorous, independent benchmarking and the potential equivalence of D-Wave’s solved tasks to efficiently solvable classical problems undermine its claims.
Will Kirkpatrick discusses the collaboration between IonQ and Ansys, and whether their touted milestone is to be taken seriously (spoiler: not really).
And a few more things from the world of atoms.
An interview with FutureHouse co-founders Sam Rodriques and Andrew White, on Asimov Press.
MIT's Raman's lab introduces STAMP, a one-step technique using 3D-printed stamps to pattern microtopography on hydrogel surfaces, enabling precise alignment of 2D contractile skeletal muscle monolayers for applications in drug screening and biohybrid robotics without requiring expensive microfabrication equipment.
Nvidia and General Motors partner on AI for vehicles and factories. Another W for Nvidia's Omniverse.
NATO Innovation fund leads a $25m series A in photonic chips company Camgraphic
Biotech
On Modality Commoditization, by Elliot Hershberg
Something here on pharma in China and why we keep writing about it.Study on Discontinuation and Reinitiation of GLP-1 Receptor Agonists
Engineered Oncolytic Virus Shows Promise in Refractory Cancer Trial
Other things we liked
Bambu Labs new H2D all-in-one 3D printer, laser cutter and plotter (If you needed ideas for our Christmas gifts).
A project to preserve memory, very three-body problem.
On pursuing the 20-mile march - slow and steady, compounding progress will always win compared to random short bursts of effort and rest.
a16z's deep dive into MCP and AI Tooling - MCP feels like just the new evolution of APIs, but everyone is going bananas about it. Surely exciting, a bit obvious - but a very good overview from a16z.
On security challenges for AI agents. arcade.dev raises start-studded seed round.
On venture manager selection and a request for funds. The folks at Nomads.vc are probably the smartest people around when it comes to allocating to venture funds. We highly recommend reading and internalizing this.
Let us know what you think of this format, and if you have anything else we should read!
