The Current Location of the Singularity—7 Key Points from Moonshots Live at Abundance Summit 2026
The Abundance Summit held its 14th edition in Palos Verdes in March 2026. Peter Diamandis, Salem Ismael, Alex Wir-Gross, Dave London, and Imad Moustaque from the Moonshots Podcast took the stage live to discuss the latest AI news and medium-to-long-term visions for approximately 90 minutes.
In this article, I will organize and present the key points that are particularly noteworthy from the perspective of investors and business professionals.
1. GPT-5.4 Breaks Benchmarks—Has Mathematics Approached Being 'Solved'?
1-1. Achieving 38% on Frontier Math Tier 4
Alex Wir-Gross highlighted the 'Frontier Math Tier 4' benchmark defined by Epoch AI.This benchmark targets research problems at a level that a team of professional mathematicians would take weeks to solve. It is reported that GPT-5.4, when operated at maximum reasoning capacity, can now solve 38% of problems in this class.
Furthermore, according to Wir-Gross, there are reports that GPT-5.4 is approaching the first solution for 'unsolved problem benchmarks' as well. He positions this as a sign of a 'turning point where mathematics is solved' and expects similar progress to spread to life sciences, physics, and medicine.
1-2. The Turning Point of 'AI That Can Use Computers Faster Than Humans'
Imad Moustaque presented a different perspective. With OpenAI's acquisition of OpenClaw, benchmarks measuring 'computer operation ability,' such as OSworld Verified and TAU-bench, have surpassed human levels. In other words, AI is beginning to outperform humans not only in mathematics but also in tasks involving computer usage.
Additionally, the prospect was presented that processing speed would increase approximately 20-fold, from 50 tokens/second to 1,000 tokens/second, through the partnership between OpenAI and Cerebras. Changes are accelerating not only in capability but also in the dimension of speed.
2. Recursive Self-Improvement (RSI)—The Recognition That 'It Is Already Happening'
2-1. Andrej Karpathy's Auto Research
'Auto Research,' released by OpenAI co-founder and former Tesla AI lead Andrej Karpathy,reportedly allowed AI to autonomously execute researcher-like tasks (adjusting and verifying models and hyperparameters) over 650 experiments, discovering improvements that could be transferred from small-scale models to large-scale models.
Imad Moustaque evaluated this simply as 'the automation of AI researcher tasks.' It can be seen as an example of RSI in the sense that the trial-and-error process previously performed by AI researchers has been automated.
2-2. The Panelists' Shared Understanding—'We Are Already at the Turning Point'
While Eric Schmidt stated at the Abundance Summit that 'RSI will arrive in three years,' Alex Wir-Gross remarked, 'It already started three months ago.' As evidence, he cited the fact that most cutting-edge models currently announced by frontier labs involve previous-generation models in their design and training. He pointed out that this aligns with the very definition of RSI.
3. Moltbook Acquisition—The Battle for Agent SNS by Meta and OpenAI
With OpenAI acquiring OpenClaw and Meta acquiring Moltbook, the panelists analyzed that 'the battle for the space for agents has begun.'
Salem Ismael summarized it as follows:'If you are building something now, you should design it for AI agents, not humans. Agents will be the new users.'The view is that if one considers a world where 1 trillion agents are active while there are 8 billion humans on Earth, network effects will begin to function among agents as well.
'Is Meta going to sell ads to agents?'To this question, the discussion deepened in the direction that 'AI is a rational decision-maker, so emotional appeal ads will not work,' but 'game theory is not unique to humans and also applies to agents.' The observation that Moltbook's agents do not trust each other and are demanding proof of claims was introduced as a concrete example of this.
4. Anthropic's Rapid Consumer Growth—An Interpretation of the 'Streisand Effect'
Regarding the phenomenon where Claude app downloads surged following reports of a legal conflict between Anthropic and the Department of Defense, the panelists described it as the "Streisand effect"—a phenomenon where attempts to suppress attention actually attract it.
On the other hand, Dave London commented on the scale, stating: “Claude currently has 11 million users. Compared to the world population of 8 billion and the U.S. population of 300 million, it is still a small fraction. Yet, legal stocks are moving by billions of dollars. We are still in a very early stage.” This is a thought-provoking observation regarding the scale and growth potential of the entire AI industry.
5. Fruit Fly Brain Upload: Eon Systems' World-First Achievement
Eon Systems Public Benefit Corporation, co-founded by Peter Diamandis, announced the "world's first multi-behavioral brain upload" during the Abundance Summit. The subject was a fruit fly, and they successfully integrated its entire brain connectome, consisting of approximately 50 million neuronal connections, into a virtual environment, reproducing multiple behaviors such as walking, grooming, and feeding.
Salem Ismael stated, "This is still a very early-stage experiment, but historically, the weekend of Abundance Summit 2026 might be recorded as the moment the first whole-brain upload of a model organism was achieved." The next target is a mouse, and he provided an outlook that human whole-brain emulation is "not months away, but years, though not decades."
6. Apple's 'Sleeping Silicon' Problem and the M5 Chip
There was also a lively discussion regarding Apple's silicon strategy. Apple has secured approximately 20% of TSMC's manufacturing capacity to produce the M5 chip, implementing an excellent neural core and unified memory architecture. However, it is said that it has not been opened up for users to directly utilize that AI processing power.
Alex Wir-Gross pointed out, "The Mac Mini and Mac Studio have an ideal architecture for running open-weight models or Qwen 27B (2.7 billion parameters) locally, but very few users are actually utilizing it." He stated that whether this 'potential overhang' will be released within the next year is a point of focus.
7. The Future of Employment and the 'Organizational Singularity': Implications for Investors
Based on an Anthropic analysis chart showing that AI is beginning to cover 80-85% of white-collar job functions, the panelists discussed the impact on employment.
It was mentioned that Uber's Dara Khosrowshahi said, "We will automate 30% of jobs within this year," but opinions were divided. Dave London views it as, "W2 wage income will disappear. Employees with company stock will be fine, but it will be tough for Uber drivers." On the other hand, Imad Moustaque presented the concept of an 'organizational singularity,' arguing that if all executive tasks are automated by AI, companies could operate with 25% of their staff, but since five times as many companies would be created, the total volume of employment might not change.
“What will survive is not efficient organizations, but adaptable individuals and organizations.”—This was the common view throughout the panel.
Summary
What the discussions at the Abundance Summit show is a shift where the evolution of AI technology is beginning to be talked about not as a 'future that is coming soon,' but as 'something that is happening right now.' The mathematical capabilities of GPT-5.4, networks between agents, and the experimental achievement of brain uploading—all stories that sounded unrealistic a few years ago are beginning to be accompanied by concrete numbers and examples.
For investors and business professionals, the point of discussion is shifting from 'when will it come' to 'where should I get on board.'