Key Highlights

  • OpenAI publishes a fresh “Our principles” framing — Sam Altman’s post reframes the lab’s stance on AGI as five guiding principles spanning safety, alignment, and stakeholder engagement. Read alongside the past week’s discourse on Claude regressions and DeepSeek’s V4 pricing volley, this looks like positioning ahead of the next round of frontier-deployment debates: capability is no longer the differentiator, governance posture is.
  • DeepSeek V4 lands its pricing punch — V4 Pro at $1.74 / $3.48 per 1M input/output tokens versus GPT-5.5’s $5/$30 and Opus 4.7’s $5/$25, with 1M-token context. Combined with Tuesday’s NVIDIA Blackwell endpoint launch, this is the open-weights camp making the long-context-agent path materially cheaper than hosted alternatives — a direct compression of the API margin pool the closed labs have been counting on.
  • Meta books 1 GW of space-beamed solar — Meta signed a capacity reservation with Overview Energy for satellite-to-ground infrared solar, targeting 24/7 data center power. Meta’s 2024 footprint was 18,000+ GWh (≈1.7M US homes); the bet is that AI training/inference demand makes orbital energy economics work despite the 2028 demonstration timeline. Energy supply is now the binding AI infra constraint, not chips.
  • Google DeepMind opens an AI Campus in Seoul — Tied to the AlphaGo decennial, the partnership opens AlphaFold, AlphaGenome, and WeatherNext to Korean researchers (AlphaFold already has 85,000+ Korean users). National-level AI partnerships are becoming the default access-and-soft-power play.
  • Theo on Markdown: “the C++ of markup languages” — A pointed argument that the language LLMs use to talk to each other (and to us) has context-sensitive grammar, multiple ways to express the same construct, ReDoS CVEs, and inline-HTML attack surface. Worth taking seriously now that markdown is the de facto agent IO format.

Analysis & Opinion

Our principles — OpenAI

OpenAI restates a five-principle framework for how it intends to develop and deploy AGI: a strategic-vision document covering safety, alignment, stakeholder engagement, and long-term considerations. The framing matters as much as the content — the labs are increasingly competing on trust posture now that DeepSeek V4 and the open-weights camp are eroding capability moats. Pair this with the recent Claude regression conversations and the AMD-exec critique of multi-agent cost structures from last week: the “responsible scaling” narrative is being repositioned as a deployment differentiator, not just an internal commitment. The principles themselves are intentionally high-level — what to watch is which decisions ship under their banner over the next quarter.

To buy this Bay Area home, you’ll need Anthropic equity — TechCrunch

A 13-acre Mill Valley property is on sale for Anthropic stock, with the seller framing it as a “diversification play” — under-indexed on AI, over-indexed on Bay Area real estate. The buyer retains 20% of upside through the lockup period. The headline is novelty; the signal is that the secondary market for frontier-lab equity is now liquid enough that pre-IPO employees can transact illiquid stock for hard assets without a tender offer. This is the same dynamic that propped up Bay Area housing during the 2010s FAANG cycle, just compressed and re-skinned for the AI era — and the kind of asset-swap creativity that tends to show up near the top of a private-market valuation regime.

New Products & Tools

DeepSeek resurfaces with cheap, capable V4 — The Rundown

V4 Pro and V4 Flash preview ship with 1M-token context and aggressive pricing — $1.74/$3.48 per 1M input/output tokens for V4 Pro, undercutting GPT-5.5 ($5/$30) and Opus 4.7 ($5/$25). External evals place V4 Pro near the top of open models on reasoning benchmarks. Reads as the consumer-facing wrapper on Tuesday’s NVIDIA Blackwell + V4 endpoint announcement.

Meta inks deal for solar power at night, beamed from space — TechCrunch

Meta signed a capacity reservation with Overview Energy for up to 1 GW of satellite-to-ground infrared solar power for its data centers. Overview’s wide-beam infrared approach (CEO claims “safe enough to stare into”) sidesteps the regulatory friction that has stalled microwave/laser-based space-solar; satellite demo targets January 2028. With Meta’s 2024 data-center load at 18,000+ GWh, this is the latest signal that AI energy procurement is moving from PPA shopping to speculative-physics moonshots.

Albertsons Media Collective brings retail signals to YouTube with Google’s Commerce Media Suite — Google

Google integrates Albertsons’ first-party shopper data into its Commerce Media Suite, enabling SKU-level sales attribution for YouTube and DV360 campaigns. Targeted at the 82% of brand-discovery journeys Google says touch its surfaces.

Research

Announcing our partnership with the Republic of Korea — Google DeepMind

DeepMind opens a Seoul AI Campus tied to the AlphaGo decennial, providing Korean researchers access to AlphaFold (85,000+ existing users), AlphaGenome, and WeatherNext for life-sciences, climate, and weather work. The “National Partnerships for AI” program continues to scale.

Interviews & Conversations

Markdown is a terrible language — Theo - t3.gg (22 min)

Theo reacts to a long-form essay arguing Markdown has degraded into “the C++ of markup languages”: a context-sensitive grammar with multiple syntaxes for the same construct, ReDoS CVEs (a real 6.9-severity in markdown-it from over-nested stars in a link), and inline-HTML escape hatches that expand the XSS surface every time a project bolts on footnotes, math, or callouts. The argument that lands hardest in the LLM context: footnotes alone push Markdown from a context-free to a context-sensitive grammar, which means agent-emitted Markdown can’t be parsed safely with the simple pipelines most of us still ship. Theo’s frank take is that he’s pushed his own T3 Chat through significant pain handling Markdown edge cases, and that the next-step solution is not MDX (too busy trying to be HTML), reStructuredText (write-hostile), or plain text (illegible) — but a purpose-built grammar with a real build system, no inline HTML, and well-defined shortcodes. Worth watching if you’re shipping anything that round-trips agent output through Markdown — which, increasingly, is everyone.

References

  1. Our principles — OpenAI, 2026-04-26 [blog]
  2. To buy this Bay Area home, you’ll need Anthropic equity — TechCrunch, 2026-04-26 [blog]
  3. DeepSeek resurfaces with cheap, capable V4 — The Rundown, 2026-04-27 [blog]
  4. Meta inks deal for solar power at night, beamed from space — TechCrunch, 2026-04-27 [blog]
  5. Albertsons Media Collective brings retail signals to YouTube with Google’s Commerce Media Suite — Google, 2026-04-27 [blog]
  6. Announcing our partnership with the Republic of Korea — Google DeepMind, 2026-04-27 [blog]
  7. Markdown is a terrible language — Theo - t3.gg, 2026-04-27 [video]