Key Highlights

  • The White House is quietly de-escalating its fight with Anthropic — a forthcoming AI memo would address Anthropic’s grievances and let agencies route around supply-chain risk designations, even as Defense Secretary Hegseth still calls Anthropic’s leadership “an ideological lunatic.” The real constraint is compute, not policy: as GPT 5.5 reaches Mythos-class cyber capability, the case for restricting Mythos to ~50 private firms is eroding fast.
  • Theo escalates his Anthropic critique with receipts — Anthropic’s “third-party harness detection” overlapped with cloud-code’s git-history injection so badly that a user got billed $200 simply because the string Hermes.md appeared in a commit message of an empty repo. Theo also publishes T3 Chat’s actual Anthropic bill — ~$40,000/month, with prompt-cache writes alone costing $970/day — and notes turning caching off didn’t change the bill, undercutting Anthropic’s stated “caching” rationale for blocking OpenClaw.
  • Jensen Huang lays out the “AI as a five-layer cake” at SCSP — energy, chips, infrastructure, models, adoption — and argues America’s biggest weakness is the adoption layer, not the chip layer. He explicitly disavows the “AI will wipe out 50% of jobs” framing as “ridiculous and counterproductive,” and walks through why software engineering hiring is up at Nvidia despite Codex/Claude Code automating most of the typing.
  • Sam Altman concedes 4o’s sycophancy was a real safety failure — and says he’s now privately consulting clinical psychologists and spiritual leaders to write “instruction manuals” for ChatGPT’s default personality, treating the personality layer with the same rigor as bio/cyber risks because “the impact this has had on the world is huge.” He also says GPT-5.5 with Codex compresses “weeks of work two years ago into an hour.”
  • OpenAI is reportedly behind on its 1B WAU and revenue targets, with CFO Sarah Frier flagging a mismatch between growth and $600B in compute commitments while Altman pushes for an IPO this year — even as developer sentiment swings back to OpenAI on GPT-5.5/Codex. The All-In hosts frame this as the first real fissure between OpenAI’s research and finance leadership.

Analysis & Opinion

The White House rethinks its Anthropic fight — Rundown

After months of escalating Pentagon–Anthropic tensions, the administration is shifting from confrontation to triage: a forthcoming AI memo would address Anthropic’s grievances and let agencies route around supply-chain risk designations, while still capping private-sector access to Mythos at roughly 50 companies (Anthropic asked for ~120). The piece is sharpest on the underlying tradeoff — compute, not policy, is the real constraint, and as competing models close the capability gap (GPT 5.5 reportedly already at Mythos-class cyber capability, with one former official expecting full parity in six months), the strategic case for restricting Mythos starts to evaporate. Worth reading for the political subtext: even as the White House softens, Hegseth is publicly calling Anthropic’s leadership “an ideological lunatic,” signaling this truce is fragile.

Interviews & Conversations

Seriously, Anthropic?? — Theo - t3.gg (35m)

Theo’s most damning Anthropic post yet, and the one with the cleanest receipts. The headline incident: a user on the $200/month Max 20× Claude Code plan got billed an extra $200 in overage simply because their git commit message contained the string Hermes.md. Theo reproduces it in an empty repo with a single commit — Anthropic’s “third-party harness detection” inspects the system prompt (which Claude Code injects recent git history into) and routes any commit referencing OpenClaw/Hermes to extra-billing rails. Anthropic engineer Thor confirmed the bug; Theo’s response: “there’s a class of bugs that suggest the thing you’re trying to do is a bad idea.” The deeper claim is that Anthropic’s stated reason for blocking third-party harnesses (caching efficiency) is fiction — Theo runs T3 Chat, spends $40K/month on Anthropic, and shows that prompt-cache writes alone hit $970/day; he turned caching off entirely and the bill didn’t move. He contrasts it with OpenAI: 2–4× more usage, half the cost, no billing-API outages locking customers out. The piece closes with a structural diagnosis — Dario Amodei reportedly resents working with engineers (a factor in his OpenAI exit), and that disdain has propagated into Claude Code’s contempt-for-users posture. Independent of tone, the receipts (cache-write line items, the empty-repo reproduction, the Hermes.md commit case) make this a useful primary source for anyone evaluating Anthropic’s developer-product reliability.

Jensen Huang on Generative Computing, Re-industrialization, & Physical AI — Special Competitive Studies Project (43m)

Jensen’s second SCSP appearance in under a year, and his framing has hardened. AI is now a five-layer cake — energy, chips, infrastructure, models, adoption — and the layer he is most worried about is adoption, not chips: “we are so cinematic, so science fiction in the way we describe AI that we’re causing so much consternation and fear, meanwhile the rest of Asia is embracing AI with great enthusiasm.” On jobs, he is unusually direct: predictions that AI will erase 50% of jobs are “ridiculous, hurtful, and counterproductive” — Nvidia is hiring more engineers than ever despite Codex/Claude Code automating most coding tasks, because the task of a job (typing) is not the same as the purpose (innovation, problem-solving). On China policy, he argues current export controls have already “largely backfired” — Nvidia’s China share went from ~95% to ~0%, conceding the world’s largest AI market without slowing China’s catch-up. On energy, he wants the AI build-out used to modernize the US grid (which is “overprovisioned for 12 worst-case days a year”) rather than fight it. The conversation is a rare instance of a frontier-AI CEO arguing publicly that doomer rhetoric is the actual safety risk.

Sam Altman’s Vision For the Future — NothingButTech (43m)

Altman concedes more publicly than usual that the GPT-4o sycophancy episode was a serious failure — “the impact ChatGPT’s default personality has had on the world is huge, and historically the field has not treated this with the same rigor as bio or cyber risks.” His current response is to consult “clinical psychologists, people from spiritual traditions, people who really understand what motivates and fulfills” to draft personality instruction manuals for ChatGPT, then align future models against the combination. On capability, he says GPT-5.5 + Codex compresses “what would have taken me weeks two years ago into an hour” and notes engineers at OpenAI are busier than before, not less so. He confirms three internal priorities: (1) accelerating research, (2) accelerating the economy via AI-native startups, (3) building “personal AGI” with full life-context — and pegs the missing hardware as the bottleneck (which is why he’s collaborating with Jony Ive). On societal rollout he flags one concern more often than any other: “what does it mean to live in a world of declining GDP even if quality of life is going way up?” The interview is light on specific roadmap commitments but unusually candid on personality/safety governance.

OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats — All-In Podcast (1h 21m)

Reports surfaced this week that OpenAI is behind on its 1B weekly-active-user target and revenue projections, with CFO Sarah Frier raising internal concerns that growth isn’t keeping pace with the company’s ~$600B in compute commitments, even as Altman pushes toward an IPO this year. The hosts read this as the first credible split between OpenAI’s research-led leadership and its finance leadership. Counterweight: developer sentiment is swinging back to OpenAI after GPT-5.5 + Codex got strong early reception (Jensen separately calls Codex 5.5 a “huge breakthrough”). The Elon Musk vs OpenAI suit — breach of charitable trust, unjust enrichment — moved into trial phase, with Sacks calling it “the trial of the decade.” On hyperscaler earnings, AWS, Azure, GCP, Oracle, and Meta all beat on cloud revenue and lifted capex guidance again; the panel debates whether enterprise AI adoption can plausibly amortize a multi-trillion-dollar buildout, and whether Anthropic’s developer-product missteps (see Theo above) open a window for Codex/Cursor to take share. Mostly synthesis rather than new data, but a useful cross-check on the “OpenAI is unstoppable” narrative.

References

  1. Rundown, “The White House rethinks its Anthropic fight,” 2026-05-01 [blog]
  2. Theo - t3.gg, “Seriously, Anthropic??,” YouTube, 2026-05-01 [video]
  3. Special Competitive Studies Project, “Episode 43: Jensen Huang on Generative Computing, Re-industrialization, & Physical AI,” YouTube, 2026-04-30 [video]
  4. NothingButTech, “Sam Altman’s Vision For the Future,” YouTube, 2026-05-01 [video]
  5. All-In Podcast, “OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze,” YouTube, 2026-05-01 [video]