Skip to content
Source · Daily Brief

AI Daily Brief — 2 April 2025

Trade-war shockwaves crashed into the AI stack on a Wednesday. Trump’s Liberation Day reciprocal tariffs sent chip stocks down, though semis were technically exempted. Underneath the macro noise: Anthropic shipped Claude for Education with three flagship universities, OpenAI dropped PaperBench, and a US-incorporated team launched Genspark — the Western counterpart to Manus.

Top stories

  • Trump declares ‘Liberation Day’ — sweeping reciprocal tariffs. 10% baseline tariff on nearly all imports plus reciprocal tariffs of 11-50% on 57 trading partners (32% Taiwan, 34% China, 20% EU, 24% Japan, 46% Vietnam). Semiconductors initially exempted from the reciprocal structure. Effective average US tariff rate jumped to its highest level since 1909. via Wikipedia
  • Chip stocks fall sharply on tariff news. NVIDIA -4.7% (deepening to 5.7% after-hours), AMD -4.5%, Broadcom -5.2%, Micron -6.4%. Wedbush flagged Taiwan (16% of NVDA FY25 revenue, ~$21B) and China ($17.1B, 13%) as the key margin/pricing risks even with the semi exemption. via Globe and Mail
  • Anthropic launches Claude for Education with three flagship partners. Tailored higher-ed version with a new ‘Learning mode’ that guides student reasoning instead of giving direct answers. Full-campus deals signed with Northeastern (~50,000 students/staff across 13 campuses), London School of Economics and Champlain College. Plus Campus Ambassadors and free-API Student Builders programs. via Anthropic
  • Genspark Super Agent launches — Western answer to Manus. Palo Alto’s MainFunc shipped a general-purpose autonomous agent built on 9 LLMs, 80+ tools and 10+ proprietary datasets. Plans/executes multi-step jobs: research, decks, spreadsheets, image/video generation, deployable web apps, even placing real phone calls with a synthetic voice. Hit $10M ARR in 9 days, $36M ARR in 45 days. via VentureBeat
  • OpenAI releases PaperBench. Agents must replicate 20 ICML 2024 Spotlight/Oral papers from scratch (understand contributions, build code, execute experiments). 8,316 individually gradable sub-tasks with rubrics co-developed with paper authors. Best agent (Claude 3.5 Sonnet New with open-source scaffolding) scored 21.0% average; top ML PhDs still outperformed all models. via OpenAI

Who shipped

Anthropic moved into higher-ed at scale. MainFunc / Genspark shipped a credible Manus-style super-agent from a US base. OpenAI dropped PaperBench as part of its Preparedness Framework — the first benchmark with an automated grader for ML replication.

Open-source pulse

No major open-weights drop dated Apr 2. The narrative around OpenAI’s tease from Tuesday continued to evolve.

Money, infra & hardware

Liberation Day announced after market close: country-by-country tariff cards (China 34%, Taiwan 32%, EU 20%, Japan 24%, Vietnam 46%). The reciprocal rates timed to take effect Apr 5 (baseline 10%) and Apr 9 (higher country rates). DCD’s analysis framed datacenter operators as quietly relieved by the semi exemption, but watching the steel/aluminum/networking-gear pass-through. via DCD

Quiet corners

No major Google, Meta or xAI launches dated Apr 2 — frontier-lab attention deferred to the tariff noise.

By the numbers

  • 10% baseline / 11-50% reciprocal across 57 trading partners
  • NVDA -4.7%, AMD -4.5%, AVGO -5.2%, MU -6.4% — first-day chip-stock damage
  • 3 — flagship universities adopting Claude for Education
  • 9 LLMs + 80+ tools — Genspark Super Agent stack
  • $10M ARR in 9 days, $36M in 45 — Genspark traction
  • 20 papers / 8,316 sub-tasks — PaperBench scope
  • 21.0% — best agent (Claude 3.5 Sonnet New) average replication score
  • Most-mentioned entity: White House

Compiled by AI Feed’s editor from verified web sources for 2 April 2025.