Recursive's $670M Bet on Self-Improving AI, Sonnet 5.5 Hits 70%, Elon Co-Leads Pentagon Push EP 299 — Key Takeaways
YouTube
Recursive's $670M Bet on Self-Improving AI, Sonnet 5.5 Hits 70%, Elon Co-Leads Pentagon Push EP 299
Peter H. Diamandis2h 26mOct 3, 2026
Watch the originalDon't buy Sonnet 5.5 — Opus 5.5 beats it on cost-versus-performance, and Gemini 4 Argon isn't competitive with either.
Key takeaways
Typesafe's Jev is a decision model for fast categorical outputs, not free text
Typesafe's Jev is a decision model for fast categorical outputs, not free text
- Answers typed questions (yes/no, category, score) in milliseconds instead of generating tokens — ideal for fraud checks and routing.
- OpenAI already launched a competing Decisions API in its Dev Day, so decision models are now a commodity.
Sonnet 5.5 beats Opus 5.5 on Terminal Bench but not on cost per attempt
Sonnet 5.5 beats Opus 5.5 on Terminal Bench but not on cost per attempt
- Sonnet 5.5 jumped from 10% to 70% on Terminal Bench 4.0, and beats Opus 5.5 at 66.4%.
- On the cost-performance frontier, Sonnet 5.5 scores lower than Opus 5.5, so there's no obvious reason to prefer it without token or latency constraints.
Router between system-one decision models and system-two reasoning models
Router between system-one decision models and system-two reasoning models
- Most enterprise micro-decisions (approve exception, route ticket, choose supplier) need fast, cheap classifiers, not frontier LLMs.
- Using an LLM for these is compared to bringing the Supreme Court to choose a checkout line.
This Dig holds 4 more insights, 4 flashcards, and 3 quotes — free with your trial.
Unlock this Dig freeStart free with 100 credits · No card, no expiry
In this video
- 1mRecursive Self-Improvement, ASI & What’s Ahead
- 5mThe Eureka Machine & AI-Driven Scientific Discovery
- 23mAI Decodes the Brain & the Future of BCIs
- 39mGene Editing, Mosquitoes & Engineering Biology
- 47mTavus, the Turing Test & AI Avatars
- 1h 0mRecursive Self-Improvement & the Road to ASI
- 1h 17mWashington Moves to Ban Recursive AI
- 1h 36mGemini 4 Argon & the Frontier Model Race
- 1h 44mSonnet 5.5 Hits 70% on Terminal Bench
- 1h 48mThe Compute Crunch & Cheaper Intelligence
- 2h 2mProject Meridian, Elon & the Future of Warfare
- 2h 9mAMA: AI, Jobs, Productivity & the Attention Economy
- 2h 24mClosing Thoughts & The Eureka Machine
“There is no realistic scenario where AI wipes out all of humanity. ... P doom is zero.”
— Richard Socher
This page is a partial, transformative summary produced by Homestake. All rights to the original content remain with its creator — please support them at the source link above.
