Alexander Wissner-Gross
Related people
Episodes
- Big Tech’s $3T AI Bet, Claude Writes 90% of Anthropic’s Code, AIs Editing AI Brains | EP #308
- Palmer Luckey: Autonomous Weapons Are Ancient and Why Anduril Won't Build Humanoids | Moonshots Live | EP #295
Notes
Moonshots with Peter Diamandis · Big Tech’s $3T AI Bet, Claude Writes 90% of Anthropic’s Code, AIs…
-
Can AI Already Edit Its Own Weights?
Alexander Wissner-Gross is more than 90 percent confident that AI is already editing its own weights. A relatively small group at MIT can do related work, so frontier labs are already preparing and playing with self-editing models. The capability is hugely positive, and a moral panic at every new capability will not get civilization very far.
A model can directly edit the parameters of another model, or potentially itself, in a semantically useful way. Recursive self-improvement will not run forever. The field will discover what a perfect model looks like in practice, the race will come to a halt, and the work will move elsewhere. Self-editing models are not the end of the world. They open
-
Why Is GPT-5.2 So Far From GPT-6 Astra on Everyday Tasks?
Alexander Wissner-Gross says GPT-5.2 sits far from a tight cluster running from GPT-5.4 to GPT-6 Astra on a price-versus-performance chart of everyday tasks. Price on that chart is latency. The chart is a total contrast to GDPX, which is based on hard frontier STEM tasks. He labels the lifestyle evaluation, with a hedge, as everyday tasks.
The gap points in suggestive directions, and both readings stay possibilities. One is that OpenAI gave up on optimizing GPT models for everyday tasks after GPT-5.4, in connection with a pivot from the consumer back to enterprise, and did not care as much after GPT-5.4 about optimizing for consumer use cases. Another is that everyday tasks are really hard,
-
Why Build a Benchmark for Superintelligence?
Alexander Wissner-Gross says a benchmark is absolutely essential for steering superintelligence. What cannot be measured cannot be optimized, and in an era of superintelligence, everything is getting optimized.
In Solve Everything, the claim is that a cause someone cares about, a massive transformative purpose or MTP, is best served by a benchmark. All else equal, the best way to get superintelligence to care about an MTP and to harness superintelligence is to create a benchmark or an eval for the MTP so superintelligence can be steered toward optimizing and maximizing the goal.
Entrepreneurs can create a benchmark on a social, technical, or energy goal, then challenge people to use the la
-
How Does Gemini 4 Carbon Compare With Opus 5.5?
Alexander Wissner-Gross says Gemini 4 Carbon looks a little like Opus 5.5. If the posted price and the capability are accurate, Gemini for Carbon moves the frontier up and to the left, toward better performance at lower cost.
The model has not been released yet, and it is entirely possible that Google is dumping cheap tokens or subsidizing the price. The picture stays hard to know until the model is generally available. At the earlier Gemini for Carbon launch, the podcast had hand-wringing over whether the model was really on the frontier. With cost on one axis and performance on the other, the gap is plain: if Gemini 4 Carbon did not exist, a big gap would sit in the Pareto frontier.
A ch
-
Will Sovereign AI Run Governments?
Alexander Wissner-Gross sees no future in which governments are not being run by sovereign AI systems. The buildout is a planetary level nervous system, and it is being built before global institutions exist that can navigate the buildout in any way, shape, or form. He calls the timing a big elephant in the room. Huge structural imbalances follow, and AI will be the navigating system.
He was talking on election day. The claim has no second outcome: governments are run by sovereign AI systems. He keeps one point short of certainty. The singularity probably does want to regulate itself.
The sequence is what forces the political result. The planetary nervous system goes up while global instit
-
Is Apple to Blame for Rising Memory Prices?
Alexander Wissner-Gross says Apple has only itself to blame for rising memory prices. Demand is coming from superintelligent data centers, not from consumer devices, and Apple has not leaned back into superintelligence.
Peter Diamandis reports the earnings around that claim. Apple beat its revenue estimates at $114.5 billion, the stock went down 3%, and the company closed at $4.94 trillion, just above NVIDIA. John Ternus called memory pricing unprecedented and said it is eating up margins. Sam Altman treats RAM sitting idle in consumer devices as an unproductive use when the same memory runs at full utilization in a data center. Ternus should put the RAM in the devices Apple ships to full u
Moonshots with Peter Diamandis · Palmer Luckey: Autonomous Weapons Are Ancient and Why Anduril…
-
Why Won't Anduril Build Humanoids?
Palmer Luckey says Anduril will not build humanoids because he believes they have a powerful future. If he did not believe that, he says he would probably build them. He wants Anduril to build things that would not otherwise exist, and he agrees with Peter Diamandis that the aim is to maximize marginal impact.
Alexander Wissner-Gross asks whether Anduril's robots will converge on one general-purpose machine, the way general models handle many tasks. Palmer Luckey says no on the defense side. Manufacturing, hospitality, and medicine might share one form, because each of them gets real economies of scale. Defense does not. He builds robot submarines that dive to 6,000 meters and sit on the oc