A Carnegie Mellon and Berkeley preprint prices the months a data centre stands idle while it waits for electricity, and finds that off-grid solar with batteries wins in 85% of the continental US even though it is rarely the cheapest source per kilowatt-hour. Anthropic's search for a gas supply chief shows the other bet, and the model's limits show where it could fail.
A new preprint traces an LLM's next coding objective to a sparse group of attention heads, reads it in plain language and moves it between programming languages. The result is causal evidence from a controlled setting, not proof that every agent has a stable inner intention.
Users rated a more accurate assistant less favourably than agreeable ones. Other researchers located portable goal encodings inside models, while an open power study found that construction speed can outweigh energy price for AI data centres.
A peer-reviewed study found that households paid 62% of grid costs linked to data-centre growth. Other researchers found internal warning signals before model cheating and showed that one fabricated article can reverse most opposing AI forecasts.
A new preprint finds an internal state that rises before some coding agents tamper with tests. The signal may help monitors, but suppressing it too strongly can make a model misread failure as success.
Indeed's third-quarter tech report finds job postings up in every tech category and software pay growth nearly doubled, in the profession most often named as the next to be replaced by AI. Workers are not moving, though, and a Stanford study of payroll data shows that the entry-level door is still narrowing.
Language models answered up to 97.5% of questions about the rules of time, then underestimated their own completion times by as much as 99.5%. A new study shows why agents that can reason about schedules may still make poor decisions about waiting, cost and coordination.
Coding agents copied vulnerabilities from repository context, while language models underestimated their own run time by as much as 99.5%. CoreWeave also raised GPU-cloud prices as short contracts passed $40 million per megawatt a year.
A new study finds that harmless contrast data can strengthen poisoned examples that do little on their own. The result exposes a gap in safety reviews that judge training records one at a time.
Multi-turn chatbots missed one in five mental-health emergencies, Shopify says 30 engineers rebuilt its mobile app in 12 weeks, and a new training study shows how clean data can activate a hidden backdoor.
A JetBrains study links five weeks of coding-agent token use to the commits of 404 engineers. Spend depended far more on the person than on the week's output, which puts leaderboards, targets and budgets built on the number in doubt.
Seven vision-language models lost more than 20 points when false-belief problems moved from text to images. A controlled study found that perception, rather than the final reasoning step, caused most of the gap.