Hyperscaler capex is now the macro story
Microsoft, Google and Amazon guided to a combined ~$210B in 2025 infrastructure spend. The second-order read: power contracts and HBM memory supply, not chips, are the binding constraint.
Fed holds, and the dot plot turns
Rates unchanged at 4.25–4.50%. Two members now see a cut by June — dovish without saying so.
OpenAI ships a cheaper reasoning tier
Priced near a tenth of the flagship, with a 128k context window. Independent evals put it close to par on math and code.
A circuit-level map of a mid-size model
Anthropic's interpretability team traces how a model routes a multi-step arithmetic task — useful when debugging odd refusals.
Agents that actually ship
Forty-five minutes on eval harnesses, why most agent demos die in production, and the tooling gap that is finally closing.
The 70B fine-tune beating a closed model on code
A reproducible community run with a surprisingly small training set. The thread digs into the data mix and what generalised.
Watching 11 sources
Yahoo FinanceHacker NewsTechCrunchMIT Technology ReviewWiredVentureBeatThe VergeAnthropic NewsOpenAI NewsRedditPodcasts
Fresh news lands every morning, and again when a story breaks.