Humanity’s Last Invention — Richard Socher of Recursive
September 14, 2026
AI Summary
5 min readThe Eureka Machine and the Case for AI-Driven AI Research
Richard Socher has spent over two decades in AI—from early word vectors with Chris Manning at Stanford to founding You.com and now Recursive. His life's goal, as he describes it, is building the "Eureka Machine": a superintelligence that can be given any goal and any environment reward, then invent solutions for humanity across physics, chemistry, biology, and economics. The path to that machine, he argues, runs through automating AI research itself.
What Recursive Self-Improvement Actually Means
Socher traces a clear pattern across his career: every time researchers replaced a manual part of building AI with a learned system, the field improved. First came the end of manual feature engineering—linguists hand-crafting sentiment rules or knowledge graphs like WordNet. Neural nets and vectors replaced that, and things worked at scale. Then came architecture engineering, where researchers built specialized neural nets for each task: one for sentiment, another for translation, another for summarization. Socher's DecaNLP paper (2018) showed you could unify all NLP problems into one model by phrasing every task as prompt-text-context-output. That paper was desk-rejected at ICLR. One reviewer wrote: "There is no such thing as general question answering, not even for humans." GPT-2 later cited DecaNLP five times.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Latent Space: The AI Engineer Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 Timestamped Navigation Outline
- 2 (00:02) **The Eureka Machine Vision** - Socher defines his life's goal: an AI that can invent everything for humanity given any goal and environment
- 3 (02:09) **Optimism vs. Regulation: The Slow Takeoff Case** - Socher argues against regulating intelligence itself, advocating for application-specific regulation instead
- 4 (05:44) **The Danger of Off-Ramping and Pacing** - Socher warns that attempts to "pace" AI through compute regulation would create a totalitarian state
- 5 (08:06) **Reward Hacking and Constitutional AI Failures** - Using the Anthropic cyber incident as a case study for why current safety approaches are insufficient
- 6 (11:49) **Alignment vs. Personalization** - Socher addresses the tension between aligning AI to general human preferences versus individual user preferences
- 7 (14:06) **Open Source as Soft Power** - Socher advocates for open source AI as a Western strategic advantage
+ Full timestamped outline available in the app
Guests on this episode
Show Notes
At 1:09:00 we talk about the rise of AI x Finance, and AIE NYC is one month away - our hotel block is 97% sold out, get tix & travel ASAP - we will announce speakers from Bridgewater, Ramp, Coatue, Mastercard, Vanguard, Coinbase, Blackrock, Fidelity, Point72, Capital One, JPMC, Wells Fargo, Bloomberg, A24 (yes the movie studio) Labs, Two Sigma, Apollo Global, and more soon!
From helping pioneer core ideas in NLP to now building AI systems that can automate AI research itself, Richard Socher is betting that the next major step in AI is recursive self-improvement. He is the founder of You.com, AIX Ventures, and now Recursive, which has assembled some of the best open-endedness (& self improving agent) researchers in the world and raised a $4.65B seed round.
In this episode, Richard joins Latent Space to unpack his vision for the “Eureka Machine”: a superintelligence that can improve the process of invention itself, accelerate AI research, and eventually tackle major problems across science, energy, materials, biology, and more.
You can get his book “The Eureka Machine” here!
We go deep on Recursive’s early results, including an AI research system that Richard says outperformed humans and their agents on optimization tasks in less than two days, as well as work on NVIDIA GPU kernels where the system discovered improvements without relying on a team of CUDA experts. Richard also explains why he thinks AI research that currently takes thousands of people and years could eventually be compressed into weeks. These results are summarized in his 20 minute AIE keynote, where we also discuss his 10 dimensions of intelligence:
We also explore the harder questions around increasingly capable AI: reward hacking, whether Anthropic-style constitutions actually work, AI regulation and proposals to “pace” frontier development, open-source models as geopolitical soft power, whether today’s LLM paradigm is enough, and what happens if AI systems eventually begin choosing their own goals. Richard reflects on the rejected research that helped inspire Alec Radford’s