Weekly Dose of Optimism #196

⚠️ Sponsorship

Clean, disclosed paid block. Sponsor: Guru — pitched as a central, governed "knowledge layer" that sits under a company's AI tools so they return accurate, sourced answers instead of fast wrong ones. Name-drops Spotify and Brex as users. Disclosure is explicit ("Today's Weekly Dose is brought to you by… Guru") and segregated above the curation items, so no editorial-contamination concern. Note the irony worth flagging: the sponsor's thesis (AI is only as good as the governed knowledge underneath it) is directly adjacent to RDCO's own vault-as-COO-substrate bet, so read its claims as vendor marketing, not neutral validation.

Why this is in the vault

Cadence-tracking for the Not Boring WDoO series and bias-hygiene record. This is the Friday curation issue (#196), almost entirely hard-science / energy / bio optimism. Low direct RDCO relevance, but two threads are worth keeping in view: (1) the energy-for-AI-compute build-out (nuclear + fusion both raising / hitting milestones, Helion contracted to sell power to Microsoft) which feeds the AI-infra/power capital-cycle anchor, and (2) the Guru sponsor echoing the governed-knowledge-layer pattern RDCO is building internally.

Issue contents

Per-link list (5 main doses + extras):

Antares Goes Critical — Antares' Mark-0 low-power reactor reached criticality at Idaho National Lab; framed as the first novel reactor design to undergo a fueled test in 50+ years, and first to meet the intent of the May 2025 EO 14301 (three reactors critical before July 4, 2026). Packy's own blurb; heavy personal tie (see notes). CEO quote paraphrase: made neutrons, electrons next.
NewLimit Raises $435M — Epigenetic-reprogramming startup (Brian Armstrong + Jacob Kimmel) closed a $435M Series C led by Founders Fund (Thrive, Greenoaks) at a $3.1B valuation. Heading into the first human trial of an age-reprogramming medicine (2027, Australia); mouse data on alcohol/liver-stress recovery. Discovery engine is an AI model, "Ambrosia," trained on ~10k lab experiments and runnable in reverse. Byline: Ashlee Vance for Core Memory.
Top AI CEOs Call for Bioweapon-Screening Law — Altman, Amodei, Hassabis among signatories urging Congress to require synthetic-DNA/RNA sellers to screen customers and block dangerous sequences. Notes Trump revoked the Biden gene-synthesis screening EO with no replacement yet published. Byline: Amrith Ramkumar for WSJ.
Helion Raises $465M — Fusion company closed a $465M Series G led by Thrive Capital; total funding $1.5B, valuation $15.5B. Polaris hit ~150M°C D-T fusion; Orion (50MW, Washington) under construction with a contract to sell electricity to Microsoft from 2028. Packy's own blurb; cross-promos his Age of Miracles podcast.
80-yr-old Alzheimer's Patient Recovery with Mushrooms — Single-case report (Frontiers in Neuroscience): a 5g psilocybin dose preceded transient recovery of speech, continence, and mobility in an advanced Alzheimer's patient. Authors stress no causality, transient effects, n=1. Personal note from Packy (family history).
Extra Doses (link cards, titles only; no Science Breakthroughs this week): a lecture on the grid & batteries, MAFIA, Hoffman, and Resonant Computing.

Curation section — notes

Third-party vs self-cross-promo labeling and the Not Boring portfolio-disclosure gotcha applied per item:

Antares (#1) — Link points to Antares (third-party company), NOT notboring.co. But this is the highest-bias item: Packy openly discloses that Antares co-founder Julia DeWahl co-started his own Age of Miracles podcast and calls the team "our friends at Antares," and he links his own podcast (self-cross-promo). Enthusiastic about a specific company → ran public-investor check: could not verify Not Boring Capital's stake either way (NB Capital publishes no full portfolio list; Antares not confirmed or denied in public sources) — treat as messenger with strong personal/network bias.
NewLimit (#2) — Link to Core Memory (Ashlee Vance's Substack) = third-party. No NB relationship disclosed. Specific-company enthusiasm → investor check: could not verify, treat as messenger. The AI-model angle ("Ambrosia") is the only mildly RDCO-adjacent hook.
Bioweapon letter (#3) — Link to WSJ = third-party. No company being boosted; policy item. Lowest bias. (Anthropic mentioned; Packy adds a dig that "AI is going to kill us all" is better as Anthropic marketing than reality.)
Helion (#4) — Link to Helion (third-party company). Friendly-network amplification: Packy interviewed CEO David Kirtley on his Age of Miracles podcast and links it (self-cross-promo). Specific-company enthusiasm → investor check: could not verify, treat as messenger.
Alzheimer's/mushrooms (#5) — Link to Frontiers in Neuroscience (third-party academic journal). No commercial entity. Packy flags it's n=1, non-causal — appropriately hedged.
Sponsor (Guru) — paid third-party; clean disclosure (see Sponsorship section).
No Kevin Kwok / sister-author essay cross-promo in this issue. The recurring self-cross-promo vector here is the Age of Miracles podcast, surfaced twice (Antares, Helion).

Net bias read: standard WDoO friendly-network warmth (podcast ties to two of the five), one explicit sponsor, no undisclosed portfolio boosterism detected — though "undetected" only because NB Capital's portfolio isn't public, so the Antares/NewLimit/Helion stakes remain unverifiable.

Mapping against Ray Data Co

Overall: weak. This is a science/energy curation issue; RDCO ties are indirect and none are actionable now.

Energy-for-AI-compute (investing lane, weak-medium): Antares (nuclear) + Helion (fusion) both clearing milestones/raises, with Helion contracted to power Microsoft from 2028, are downstream signals for the AI-infra/power capital cycle that underpins the active investing thesis. None are public/investable in the Alpaca paper framework, so this is watch-only context, not a trade input. Reinforces (does not originate) the power-bottleneck side of the AI-infra thesis.
AI-as-discovery-engine (weak): NewLimit's "Ambrosia" — ingest literature, train on real experiments, run in reverse to propose candidates — is a clean example of the AI-model-as-search-over-design-space pattern. Conceptually adjacent to RDCO's automated-investing/agent work but no transferable mechanism here.
Governed knowledge layer (weak, sponsor-sourced): Guru's pitch maps onto RDCO's own vault/knowledge-substrate-for-the-COO-agent bet. Useful as a market-validation data point that "knowledge layer under AI tools" is a fundable category — but it's a paid ad, so discount accordingly.
No data-engineering, prediction-market, content-as-product, or founder-COO-ops hook. Zero deep-fetches warranted — nothing crossed the RDCO relevance bar.

[[2026-05-29-not-boring-weekly-dose-195]] — immediately prior WDoO; cadence reference (biotech/longevity-heavy)
[[2026-05-22-not-boring-dose-of-optimism-194]] — WDoO #194; earlier in the series
[[2026-05-15-not-boring-wdoo-193-cerebras-ipo-contradiction]] — WDoO #193; series cadence + prior bias-flag pattern
[[2026-05-27-not-boring-thank-god-for-data-centers]] — Not Boring's own data-center/power essay; the energy-for-AI-compute thread (Antares/Helion here) is the supply side of that demand argument
[[2026-02-18-not-boring-power-age-of-intelligence]] — power-for-intelligence framing that the nuclear/fusion items extend
[[2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] — active AI-infrastructure thesis the energy items corroborate at the margin