Key Highlights

  • OpenAI shipped GPT-Realtime-2, a voice model that finally closes the reasoning gap. Big Bench Audio jumped from 81.4% (predecessor) to 96.6%, and the new model can call tools simultaneously, reason mid-utterance, and “talk while it thinks.” A live translator covering 70+ languages and a streaming-transcription model shipped alongside. Zillow, Priceline, and Deutsche Telekom are already in production. The structural read: voice agents are graduating from turn-based dialog to multi-step workflow execution — the same arc text agents made in 2024-25, compressed into 18 months.
  • Karp, Jensen, and Dario all showed up this week with sharply different theories of where the AI bottleneck is, and they don’t agree. Karp’s Q1 print (100% US growth, Rule of 145, free cash flow this quarter > revenue from same quarter a year ago) was the loudest argument that AI without an ontology is theater — he spent 15 minutes on the earnings call calling competitors “AI slop” and saying the demos work but the deployments don’t. Jensen at Milken made the opposite case: capacity is the bottleneck (agentic AI is 1000× more compute than generative AI), not platform discipline; both OpenAI and Anthropic just turned gross-margin-positive in the last 3-6 months and “are racing for capacity” because the unit economics finally work. Dario at JPMorgan added the 6-12 month estimate for Chinese open-weight models to catch frontier US labs and predicted individual SaaS companies will go bankrupt as moats collapse — useful to read in tension with Martin Alderson’s argument (covered below) that open-weight licensing is tightening, not loosening.
  • Theo’s “What’s next?” video is the most useful single audit of GitHub alternatives anyone’s published this cycle. His framework: GitHub is dying, GitLab and Bitbucket are Gen-2 alternatives that are “just worse GitHub,” and the only mature open option worth recommending today is Forgejo / Codeberg (a community fork after Gitea went private). He went on-camera and donated $1,200 + $400/month live; that’s the credibility he’s putting behind it. The Gen-3 piece is Pierre’s code.sto (an ultra-low-latency Git cloud built for agent throughput — they hit 15,000 repos/min for 3 hours straight while GitHub buckles at half their volume) plus Entire (the new $60M-seed company from GitHub’s last CEO, building durable agent-context history alongside Git) and Zed’s Delta DB (CRDT-based realtime collab). The point worth filing: the Gen-2-to-Gen-3 jump may leave Git itself behind, the way Gen-1-to-Gen-2 left SVN behind.

Analysis & Opinion

Open Weights Are Quietly Closing Up — and That’s a Problem — via Lobsters

Martin Alderson argues open-weight LLMs from DeepSeek, Qwen, and others are functionally the generic-pharma price ceiling of the AI economy: they cap pricing power on closed frontier models because “if frontier labs raised prices 5× overnight, a huge amount of people would just switch.” The trend he documents is that this constraint is quietly weakening — Meta has stopped releases, Alibaba is moving more weights behind API-only access, and Mistral is layering commercial-use restrictions onto licenses. That’s directly opposite to the “abundant defender swarms” framing Jensen used at Milken (where open source is the cybersecurity dome against frontier-model attackers), so the two pieces are worth reading against each other.

New Products & Tools

OpenAI closes reasoning gap in voice agents — The Rundown

Three new voice API models — GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper — push voice agents past the turn-based-chatbot plateau. Realtime-2 brings GPT-5-class reasoning, can use tools simultaneously instead of sequentially, and supports “talk while it thinks” interaction with markedly better tonal realism. Big Bench Audio scores: 96.6% vs 81.4% previously. The accompanying live translator covers 70+ languages with streaming transcription. Zillow (real estate agents), Priceline (travel concierge), and Deutsche Telekom (CX) are first-wave production deployments — same pattern as text agents in 2024, with a roughly 12-18 month lag.

Interviews & Conversations

Alex Karp Leaves Audience Speechless — David Carbutt (19m, Palantir Q1 earnings reaction)

Karp’s earnings-call rant is the cleanest articulation of the “ontology vs. slop” thesis on record. Numbers first: 100% US growth, Rule of 40 at 145, Q1 free cash flow ($791M on $1.4B revenue, 56% margin) larger than total revenue from the same quarter a year ago, and a sales force of “70 people, only 7 of whom actually sell” doing the work of a normal company’s 7,000. His core argument: agentic AI demos look identical from the outside whether they’re on a real ontology or not, but in production deployment only the ones with a structured object/permission/relationship layer actually work — and Foundry + FDEs + ontology is the only such stack shipping at scale right now. Net dollar retention 139%, biggest problem is meeting demand. The macro read is that Karp’s framing is the most aggressive available case that the platform layer, not the model layer, is where AI economics actually accrue — a direct counterargument to the “GPT wrapper” thesis. Worth reading against Jensen’s view that the model labs themselves just turned gross-margin-positive.

Leading in the Age of AI: Jensen Huang at Milken Global Conference 2026 — Milken Institute (46m)

Jensen’s clearest public articulation yet of the “five-layer cake” framing — energy, chips, infrastructure, models, applications — and which layer is most constrained at any given moment. Headline numbers: agentic AI requires ~1000× more compute than generative AI, demand is up another 100× on top of that as user counts grow, and “GPUs we sold four or five years ago are rising in price faster than good wine.” His argument that OpenAI and Anthropic both turned gross-margin-positive in the last 3-6 months is the most consequential disclosure in the conversation — that’s why both companies are now “just racing for capacity” rather than burning cash to prove product-market fit. On safety: he’s a self-described pragmatist (not boomer or doomer), thinks Hinton is “completely wrong that smart people aren’t already working to prevent” the worst outcomes, and his cybersecurity proposal is “swarms of white blood cells” — open-source models like Mythos deployed as defenders, beating offense via abundance not parity. On jobs: the radiologist parable — “computer scientists kept saying that job would be eliminated, instead radiology departments became the most profitable units in hospitals because radiologists could see more patients” — is his standard rebuttal to the 20-30% unemployment forecasts. Quietly he also said Nvidia is probably done investing directly in OpenAI/Anthropic (“they don’t need it anymore”) and is instead looking at energy and infrastructure layers where $1 of Nvidia capital can unlock $100 of ecosystem investment.

What’s next? — Theo - t3.gg (50m)

Comprehensive post-GitHub-meltdown audit of every realistic alternative, with strong prescriptive opinions. His generation-of-product framework: SVN → Gen-1; GitHub / GitLab / Bitbucket → Gen-2; whatever’s coming next is Gen-3 and probably leaves Git behind entirely. GitLab and Bitbucket are dismissed — GitLab as “a worse GitHub the same way Azure is a worse AWS” (528K commits of Ruby slop, UX disasters in releases/history/code-review), Bitbucket as “Git solutions for teams using Jira” with a marketing page that mentions Jira five times. Forgejo / Codeberg gets the strongest endorsement of the video — democratically-governed nonprofit, written in Go (not Ruby), 25K commits not 500K, releases tab works correctly, supports push-to-create, has push-based agent writes for free; he donated $1,200 + $400/month live on stream and changed his planned video direction to recommend it as the immediate move. The Gen-3 candidates are three companies betting on different theories: Pierre / code.sto is rebuilding Git’s storage layer for agent-throughput workloads (15,000 repos/min sustained for 3 hours, vs GitHub buckling at half that volume) plus open-source primitives like diffs.com and trees.software; Entire ($60M seed, ex-GitHub CEO Thomas) is building durable agent-context history that lives alongside Git so agents know why code changed not just what changed; Zed’s Delta DB uses CRDTs for realtime collab that interoperates with Git but isn’t bound by snapshot semantics. Theo’s emotional close — that the cost of leaving GitHub isn’t technical but social, because “you can’t click someone’s profile and see 20 years of who they are anywhere else” — is the line worth quoting if you’re trying to explain why this transition is harder than people think.

Dario Amodei and Jamie Dimon on AI boom, regulation & jobs — CNBC Television (5h panel — clip)

Three load-bearing claims from Dario worth filing: (1) Chinese AI models will catch frontier US labs in 6-12 months, narrowing the safety-alignment window correspondingly; (2) individual SaaS companies will go bankrupt as their moats evaporate, but only the ones who don’t pivot — there’s still room for incumbents who recognize what’s happening; (3) on regulation, he prefers an NTSB/post-deployment monitoring model over an FDA pre-approval model — explicitly rejecting Becky Quick’s framing that AI should be vetted before release (“the FDA slows down medical progress a lot”), and instead arguing for “some oversight while it’s out there.” Dimon’s frame on the $1T+ AI capex: “in total it will make sense; if you want to pick winners and losers you’ll have a hard time.” On jobs, both pushed back on the Verizon CEO’s recent 20-30% unemployment forecast within 2-5 years — Dimon arguing capitalist economies have absorbed every prior automation wave (agriculture, electricity, internet), Dario more circumspect that “this one is faster.” The Pentagon-Anthropic conversation continues; Anthropic’s Mythos model is the negotiation centerpiece. Notable cross-reference with Jensen, who in the Milken interview praised Anthropic specifically and said “I hope the US government and Anthropic work it out” while making clear Nvidia takes no position on whether US technology should be deployed by US military.

Elon Musk Gets Confronted In Interview But FIRES Back — The Money Investing (45m, advertising-festival interview)

Wide-ranging conversation at an ad-industry festival that’s mostly familiar Musk positions, with three threads worth flagging. AI risk: he echoes Hinton’s 10-20% extinction probability (“but the glass is 80% full”) and predicts the most likely outcome is “universal high income, work optional” — but warns of a coming “crisis of meaning” if AI can do everything humans can do but better. AI-native ad targeting: X has moved to a fully AI-vector-space matching system where user posts and ads occupy the same vector space and get correlated; he claims this is structurally different from old-Twitter’s brand-only model and credits TikTok with pioneering it. Search disruption: he expects “significant disruption in internet search” because “if AI can give you a better answer than a bunch of links, you’ll prefer that over Google.” On Optimus: “20 billion-ish humanoid robots eventually,” personalized with snap-on plastic shells, will be regarded as friends. On longevity: he’s not investing in life-extension because “if leadership never dies, society ossifies — a lot of people don’t change their minds, they just die.” Useful as a contrast piece against Jensen, who at Milken explicitly rejected the doomer framing Musk endorses here.

References

  1. The Rundown, “OpenAI closes reasoning gap in voice agents,” 2026-05-08 [blog]
  2. Martin Alderson, “Open Weights Are Quietly Closing Up — and That’s a Problem,” Lobsters, 2026-05-06 [blog]
  3. David Carbutt, “Alex Karp Leaves Audience Speechless,” YouTube, 2026-05-05 [video]
  4. Milken Institute, “Leading in the Age of AI: A Conversation with NVIDIA CEO Jensen Huang,” YouTube, 2026-05-05 [video]
  5. Theo - t3.gg, “What’s next?,” YouTube, 2026-05-06 [video]
  6. CNBC Television, “Anthropic’s Dario Amodei and JPMorgan’s Jamie Dimon on AI boom, AI regulation & impact on jobs,” YouTube, 2026-05-06 [video]
  7. The Money Investing, “1 Hour Ago: Elon Musk Gets Seriously Confronted In Interview But FIRES Back!,” YouTube, 2026-05-06 [video]