Ilya Sutskever's SSI Inc raises $1B
A new AI startup co-founded by former OpenAI chief scientist Ilya Sutskever has raised $1B at a reported ~$5B valuation to pursue “safe superintelligence,” despite having no product yet. Commenters debate whether this represents rational backing of a uniquely strong team in a capital‑intensive race for AGI, or a late‑stage bubble driven by hype, FOMO, and VC portfolio math, with NVIDIA and cloud providers as the most reliable near‑term winners. Many are skeptical that scaling current LLM architectures can deliver true AGI or safety, while others argue that even incremental advances toward more capable, reliable models could justify enormous valuations if they become core infrastructure across the economy.
Scale of the round and valuation
- Many are stunned: $1B in cash at ~$5B valuation for a months‑old, pre‑product, pre‑revenue company is seen as “insane” or historically large for a seed.
- Defenders argue this is a capital‑intensive space (GPU, training costs), so a huge seed is rational, especially when betting on a top-tier team.
- Some note that “$1B isn’t even competitive” at the frontier, while others think it’s plenty to do several large training runs and assemble a world‑class team.
Hype, bubble, and VC dynamics
- Strong disagreement on whether this signals peak AI hype or a still‑escalating bubble.
- Comparisons to dot‑com era (Amazon vs Webvan) and to crypto: some see smart frontier investment, others see “degenerate gambling” and musical‑chairs exits.
- Several comments note big VCs may care as much about markups and raising the next fund as about ultimate business viability.
AGI / superintelligence feasibility
- Intense debate over whether scaling current transformer LLMs can reach AGI or superintelligence:
- Pro‑scaling side: next‑token prediction plus larger models and better data implicitly learns world models; current systems already show broad, general capabilities.
- Skeptical side: transformers have architectural limits (fixed depth, weak long‑term learning, poor reasoning, counting, OOD generalization); may need new paradigms.
- Some argue we don’t even have a clear, testable definition of “AGI” or “superintelligence,” making timelines and milestones unclear.
“Safe” superintelligence and alignment
- Confusion over what “safe” means:
- One view: “aligned with the controller’s goals” (potentially attractive to states/authoritarians).
- Another: “won’t harm or extinct humanity,” which many doubt is technically achievable.
- Skeptics worry safety talk is mostly branding or will erode under profit pressure, as “Open” did for OpenAI.
- Others see a real market for predictable, non‑hallucinating, liability‑bounded systems in regulated sectors (healthcare, law, finance, government).
Data, compute, and ecosystem
- Expectation that a significant fraction of the $1B goes straight to GPUs or cloud credits; good for Nvidia.
- Concerns about diminishing access to high‑quality training data (APIs locking down, copyright, lawsuits), making it harder for late entrants.
- Some see this as a “Manhattan Project”‑style bet; others say that’s inapt because we lack a known, well‑founded path to AGI.