AI May Threaten Widely Used Cryptography Within Two Years, Scott Aaronson Warns

·2 views

Former OpenAI safety researcher and quantum computing scholar Scott Aaronson discusses AI's growing ability to solve hard mathematical problems. He notes that OpenAI has publicized nearly 400 AI-solved results in mathematics and theoretical computer science, but none in cryptography, which he attributes to a deliberately cautious approach. Aaronson argues it is plausible that AI models could break one or more widely used encryption algorithms within one to two years, a disclosure that would need to reach the entire world rather than a single company. The interview also highlights how AI-generated proofs, such as one resolving the Unique Games Conjecture, are difficult for humans to read, and how companies like OpenAI and Anthropic are taking different approaches to publishing and explaining such results. Aaronson concludes that the central challenge is preserving human mathematical understanding and academic communities in an era where anyone can query AI for answers to open problems.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Silent Gaps in Billing Recovery Drive Customers Away, Argues DEV Community Post

A DEV Community piece argues that the silence between automated payment-recovery messages leaves customers guessing whether a failed payment was fixed or their subscription is ending. Such gaps tend to be read as indifference or a trap rather than patience. The author recommends clear confirmations when a card is updated, plain notices when a retry succeeds, and a closing message stating what happens next when recovery ends. The post notes that Swipe Again builds these confirmations and closures into its recovery flow as standard sends. The author concludes that customers tolerate payment failures but tend to cancel when left to guess.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Phantom Reads From Misconfigured Isolation Levels Caused Intermittent Revenue Mismatches

An engineering team found that category revenue totals sometimes disagreed with logs recorded moments earlier, but only during peak production batch runs. Investigation traced the cause to a mismatch between database transaction isolation levels and the default session behavior of their SQLAlchemy ORM. Under Read Committed isolation, a reader transaction that repeated a range query could see rows newly committed by a concurrent writer, causing sums and counts to shift mid-transaction. To study the problem, the team built a deterministic Python simulation that runs concurrent readers and writers across different isolation levels and logs phantom read occurrences. The write-up presents the simulation as a way to show how stricter isolation settings such as Serializable prevent the anomaly.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Recovery Software Must Serve Both Merchants and Their Customers

Recovery software has two audiences: the merchant who signs the contract and pays, and the merchant's customer who experiences the emails and update links. The author argues that every surface, including dunning emails, save offers, and reports, is read by both groups, so designing for only one risks losing the other. The customer's impression of how a business handles problems can determine whether they stay. The author's product, Swipe Again, uses the merchant's branding so the customer never encounters the software directly. Success, in this view, means the subscriber feels the company handled a payment hiccup gracefully, with the recovery machinery kept invisible.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developers Urged to Keep Understanding Code as AI Writes More of It

AI tools can now generate functioning applications from plain-language descriptions, letting even inexperienced users produce working software in minutes. The article questions whether this speed makes developers more skilled or merely better at prompting AI for code. It argues that producing code quickly does not guarantee the understanding needed to maintain and debug it. The piece frames the goal as using AI to build capability rather than dependence, distinguishing unnecessary frustration from struggle that builds real engineering knowledge. It also highlights a growing risk of accumulated "understanding debt" when developers ship code they cannot fully explain.

0 comments Read more at DEV Community