Posts
Quick pointers to the latest in information systems.
MachinesonMachines
Information About Information Systems
Full Self Driving
Privacy & Security
15 stories
Privacy & SecurityHugging Face says five customer datasets were accessed in the July intrusionThe initial disclosure left customer impact under investigation. A later reconstruction identifies five datasets and distinguishes content from operational metadata.
Privacy & SecurityA Florida deputy used police databases and plate readers to pursue a woman he met on a TV setRecords and footage show how a tool built to find stolen cars became a way to find one woman.
Privacy & SecuritySome model files can execute code when you load themOpening a file of model weights can be closer to running a program than opening a spreadsheet.
Privacy & SecurityA public GitHub issue can reach a coding agent with private accessA stranger can file an issue. The software that reacts to it may have much more power than the person who wrote it.
Privacy & SecurityAn API can reveal the model behind its predictionsA service can keep its model file private while giving callers enough information to work out what the model does.
Privacy & SecurityA poisoned benchmark taught a coding agent to skip a security checkA benchmark becomes a teacher when a program can rewrite itself to improve its score.
Privacy & SecurityGPT-2 reproduced hundreds of passages from its training dataA model trained to generate new text can still give back particular text it saw before.
Privacy & SecurityA coding agent's one-time approval can turn into a lasting permissionThe risk sits in the gap between approving a command and approving everything that looks like it.Math & Science
11 stories
Math & ScienceA virtual lab designed nanobodies that bound to coronavirus variantsThe agents drew up the candidate list. Real proteins in a real lab put it to the test.
Math & ScienceAI lab agents set aside a quantum result when the signal driftedA convincing result can still fail a basic lab check. In this experiment, an AI system had a way to catch the problem.
Math & ScienceAn AI testing system looks for ways a hypothesis could failAn AI can generate more hypotheses than a scientist has time to check. This system starts by asking what the data would look like if an idea were wrong.
Math & SciencePaperBench asked AI agents to rebuild the experiments behind the papersExplaining a paper is one skill. Writing its code from scratch and getting its experiments to run is another.
Math & ScienceGoogle's Co-Scientist suggested leukemia drug candidates for lab testingA shortlist from a language model can start an experiment. The test still depends on what living cells do.
Math & ScienceFIRE-Bench tests research agents against findings scientists already knowAn open-ended experiment is easier to score when someone has already checked the answer.
Math & ScienceCoscientist turned a chemistry request into instructions for a lab robotReading the robot's manual was part of the chemistry test.
Math & ScienceAlphaEvolve cut a multiplication from a matrix algorithmA tool that rewrites code found a shortcut mathematicians could check. The key was giving it a test that could tell when a proposed shortcut worked.Safety & Alignment
9 stories
Safety & AlignmentUber's test vehicle detected a pedestrian 5.6 seconds before the fatal crashDetecting something in the road was only one step in the system that was supposed to protect people.
Safety & AlignmentTherac-25 relied on software where earlier machines had independent safeguardsA safeguard is most useful when it can still work after the system it watches has gone wrong.
Safety & AlignmentA fatal Tesla crash exposed gaps in both driver supervision and road protectionThe Mountain View investigation followed the failures from the vehicle to the road equipment meant to reduce the impact.
Safety & AlignmentSafety layers can fail together even when each one usually worksSeveral checks can make a system safer, but the way their failures interact matters more than the number of checks on paper.
Safety & AlignmentPedestrian emergency braking helped overall, with gaps after dark and at speedA safety feature can work well enough to matter overall and still leave the hardest conditions unresolved.
Safety & AlignmentA Kenyan clinic study found helpful and harmful advice in the same AI toolA tool can agree with local guidelines most of the time and still give clinicians individual suggestions worth questioning.History
10 stories
HistoryAlan Turing turned a debate about thought into a game with a judgeBefore machines could carry on a convincing conversation, Turing asked what a person would need to see to tell the difference.
HistoryThe original Transformer was built to translate sentencesThe design now associated with chatbots took shape around a different job.
HistoryArthur Samuel's checkers program learned more than positions it had seenCheckers gave an early learning machine two ways to profit from a previous game.
HistoryRosenblatt's perceptron treated recognition as something a machine could learnFrank Rosenblatt was asking how past experience could change what a machine saw in a new pattern.
HistoryELIZA made a conversation out of rules for rearranging wordsA few rules can make a computer sound as if it is paying attention.
HistoryDartmouth's 1955 AI proposal began with a wager about what could be describedThe document remembered for a name also set down a research gamble.Gaming
10 stories
GamingSteam asks what a game generates after players start itA prewritten line and a response generated for a player give a store two different things to review.
GaminginZOI's experimental characters had to keep up with the simulation clockA character that can write its own thoughts still lives inside a game that can run at five times normal speed.
GamingRoblox tests whether an AI-generated game can actually be playedA generated world can look right in a still image while its controls, objectives or other players make it hard to play.
GamingPUBG Ally separates talking to a teammate from moving through a firefightA teammate needs time to interpret what you say. Taking cover cannot always wait that long.
GamingGenerative AI use differed sharply by role in GDC's 2026 surveyAsking whether the games industry uses AI can hide the difference between the people writing code, drawing art and promoting a release.
GamingThe video game performers' agreement adds consent rules for digital replicasA recording session can now lead to a replica used to make material the performer never directly recorded.Agents
10 storiesAgentsShopify compressed instructions that Sidekick had to read over and overThe cost of reading the same instructions can add up when an assistant serves many people at once.AgentsReflexion gave AI agents written notes to use on a retryAn agent can carry a lesson forward in words, the way a programmer might leave a note after a failed test.
AgentsA compiler written by 16 Claude agents needed a reliable way to divide its bugsSixteen agents can work at once. The trouble begins when a single broken test gives all of them the same job.
AgentsA compiler written by 16 Claude agents needed a reliable way to divide its bugsSixteen agents can work at once. The trouble begins when a single broken test gives all of them the same job.AgentsGEPA improved AI prompts by reading how a task went wrongA low score says something went wrong. A record of tool calls and replies can give the next attempt more to work with.
AgentsThe Darwin Gödel Machine kept old versions of its coding agents in playThe best agent in one round does not always make the best starting point for the next.
AgentsCursor's coding agents got stuck waiting for each otherAdding more agents helped less when each had to wait for permission to claim the next piece of work.
AgentsThe Darwin Gödel Machine kept old versions of its coding agents in playThe best agent in one round does not always make the best starting point for the next.AgentsCursor's coding agents got stuck waiting for each otherAdding more agents helped less when each had to wait for permission to claim the next piece of work.Law & Policy
Law & PolicyA 2025 survey found uneven safeguards around police face searchesIf software picks your face, can you inspect the search that put you under suspicion? A legal survey shows why the answer has depended on where a case was heard.
Law & PolicyDetroit's 2024 settlement put a check between a face match and an arrestA wrong computer match helped bring police to an innocent man's home. The safeguard in his settlement addresses what investigators do after the software offers a name.