‘He’s a loser’: Canada's Ford hits back at Trump

When passing tests still let a software bug slip through undetected

A developer maintaining qbofile, a browser-based accounting file converter, discovered that green unit tests can mask incorrect logic when both sides of an assertion are generated by the same flawed code. The bug involved converting transaction split amounts from IIF format, used by QuickBooks Desktop, to QIF format, read by Quicken and GnuCash, where a sign convention error would cause both sides of the test equation to flip simultaneously, keeping tests green. Mutation testing confirmed the assertions were wired to the code but could not determine whether the underlying logic was correct. The developer resolved the uncertainty by building a purpose-designed QIF file and importing it into GnuCash, verifying the split totals were accurate — turning a confident assumption into a confirmed measurement. Separately, a real defect was uncovered during the review process, unrelated to signs and involving how IIF payee and memo fields map to QIF's distinct P and M fields.

AWS Shared Responsibility Model: What Cloud Security Falls on You vs. AWS

The AWS Shared Responsibility Model divides cloud security into two distinct domains: AWS secures the underlying infrastructure, while customers are responsible for what they deploy and configure on top of it. AWS covers physical data centers, hardware lifecycle, hypervisor isolation, global network security, and patching of managed services like RDS and S3. Customers are accountable for data encryption, identity and access management, OS patching on IaaS instances, network configuration, and credential management. The boundary between these responsibilities shifts depending on the service model used — IaaS, PaaS, or SaaS — meaning there is no single fixed line applicable to all scenarios. Most real-world cloud security incidents stem not from AWS failures but from customers mistakenly assuming AWS handles tasks that are actually their own responsibility.

Two UP Children Go Missing After Attempting YouTube Secret House Challenge

Two children in Uttar Pradesh left their home after being inspired by a 'Secret House Challenge' trend on YouTube. The incident prompted a swift response from local police and the state's Mission Shakti team. Authorities launched a search operation and successfully located both children within approximately two hours. The children were safely recovered and reunited with their respective families.

Iran Tells US Coercion Won't Work as Pezeshkian Meets Pakistan Army Chief

Iranian President Masoud Pezeshkian held talks with Pakistan Army Chief Asim Munir in Tehran amid rising regional tensions and ongoing US sanctions. Pezeshkian called on the United States to reconsider its approach toward Iran, warning that pressure tactics obstruct diplomatic progress. He stressed that coercive methods make meaningful negotiations more difficult rather than productive. Both leaders underscored the importance of bilateral cooperation in promoting peace and stability across the region.

Developer disables semantic cache after tests reveal it returns opposite answers with full confidence

A developer building a regulatory document retrieval system implemented a semantic cache using cosine similarity to avoid redundant model calls, setting a threshold of 0.92 to match near-identical questions. After a colleague challenged the safety of that threshold, a six-day investigation revealed the cache could return answers to questions with opposite meanings — such as 'patients with fever' versus 'patients without fever' — without any warning signal. Testing with 20 handcrafted Spanish administrative question pairs showed the highest adversarial similarity reached 0.9984, while the lowest genuine paraphrase scored only 0.7470. This meant no threshold could safely distinguish opposite-meaning questions from true paraphrases, making the cache fundamentally unsafe for rule-based queries where citation accuracy is critical. The developer disabled the cache by default and published a full write-up, noting that a bad cache hit is more dangerous than a bad search ranking because it bypasses the model entirely and shows no sign of failure.

Why Backtests Lie: The Overfitting Traps Every Quant Trader Must Avoid

In quantitative trading, a profitable backtest is easy to produce but often meaningless, as strategies tuned on historical data tend to memorize noise rather than capture genuine market patterns — a problem known as overfitting. Key culprits include multiple testing, where running thousands of parameter variations almost guarantees a winner by chance, and lookahead bias, where future data inadvertently leaks into past decision points. Survivorship bias further distorts results when backtests exclude delisted or bankrupt assets, while ignoring realistic transaction costs and slippage can make a losing strategy appear highly profitable. Experts recommend walk-forward testing on rolling out-of-sample windows, keeping a locked holdout dataset touched only once, and favouring strategies with fewer parameters that perform robustly across a range of settings. Accounting honestly for the total number of strategies tested — using tools like the deflated Sharpe ratio — is essential to distinguishing a genuine edge from a statistical accident.

Delhi-NCR weather today (August 25): IMD forecast

How Universities Should Prepare Founders

Article URL: https://paulgraham.com/prepare.html Comments URL: https://news.ycombinator.com/item?id=49428121 Points: 8 # Comments: 0

Flood-Hit MP Teachers to Get Alternative Classrooms After Viral Video

Teachers and students in Bhind, Madhya Pradesh, have been risking their lives crossing a flooded river to reach school during the monsoon season. A video highlighting the dangerous situation went viral, prompting swift action from local authorities. Temporary classrooms have since been arranged to ensure continuity of education without endangering lives. Additionally, officials have confirmed that teachers will not face penalties for missing electronic attendance under these circumstances.

Developer's AI agent silently corrupted its own 13KB memory file into 49MB over three days

A developer's autonomous agent, designed to wake every four hours and log its state to a file called STATUS.md, silently inflated that file from 13KB to nearly 49MB over three days due to a subtle Python bug. The flaw stemmed from inverted string indices causing a slice to return an empty string, which was then passed to Python's replace() method — inserting new content between every single character in the file. Because the corruption added rather than removed data, routine checks using grep and sed all passed, masking the problem entirely. The developer concluded that checking for content presence is structurally blind to dilution-style failures, and that coarse signals like file size can catch errors that precise content checks miss. As a fix, post-write size assertions, index-order validation, and structural checks on section headings were added to the agent's workflow.

The CIA's Secret Funding of Modern Art During the Cold War

Declassified documents and historical research have revealed that the CIA covertly funded and promoted Abstract Expressionism and other modern art movements during the Cold War era. The agency operated through front organizations to channel money to artists and exhibitions, aiming to position American culture as a symbol of freedom against Soviet ideology. The program, known as the Congress for Cultural Freedom, supported prominent artists without their knowledge in many cases. Historians debate whether this constitutes a 'psy-op' or simply a strategic cultural diplomacy effort. The revelation raises ongoing questions about the intersection of government influence and artistic independence.

Mumbai food safety raid leads to destruction of over 4,700 litres of suspect milk

Mumbai food safety officials conducted raids across multiple establishments as part of a public health crackdown. Authorities seized and destroyed 4,710 litres of milk found to be adulterated or otherwise suspect. The confiscated milk was valued at approximately Rs 1,85,780. The action was taken to safeguard consumers from potentially harmful dairy products.

No one has built a fully autonomous LLM pipeline that self-approves content without human review

Researchers surveying existing LLM publishing pipelines found that every known system either skips content-quality checks entirely or halts before publishing to await human approval. Some pipelines automate technical checks like build compilation or link validation but cannot assess whether written content meets safety or policy standards. Others perform deeper content review — such as tone analysis or code execution — yet still require a human sign-off before anything goes live. One team did allow an LLM to publish autonomously from the start, but only added an automated check after readers flagged broken code examples in published posts. The gap identified is a specific untested design: an LLM drafts content, a fixed non-LLM ruleset evaluates it, and the system publishes or withholds the post with no human involved at any stage.

Enterprises May Soon Adopt Shared AI Harnesses to Replace Fragmented AI Tools

A growing pattern in large enterprises involves teams independently building separate AI applications, each duplicating infrastructure like authentication, logging, and policy checks. This fragmented approach leads to dozens of overlapping tools with inconsistent security assumptions, making it difficult to manage at scale. The proposed solution is a shared enterprise AI harness — a configurable open-core platform that standardizes identity, model access, governance, and audit requirements across the organization. Rather than replacing individual applications, the harness acts as a common foundation layer that makes each app thinner and more secure by handling the complex governance work centrally. This model mirrors how enterprises already standardize cloud platforms and API gateways, with plugins allowing each company to tailor the harness to its own internal systems and policies.

Are banks open or closed today? Check here

US Set to Revoke B1/B2 Visas of Up to 200,000 Asylum Seekers

The US State Department is preparing to announce the cancellation of B1 and B2 visas held by individuals who have applied for or are currently seeking asylum. The move is expected to be formally announced within the coming weeks. The revocation would target visas issued between 2016 and 2026 whose holders pursued asylum claims. The policy could affect up to 200,000 foreign nationals currently in the United States.

Trump gets a win on mail-in voting, for now

Karnataka suspends Amazon, Instamart, BigBasket licences over toxic datura sales

Karnataka authorities have suspended food licences of Amazon, Swiggy Instamart, and BigBasket after toxic datura fruits and seeds were found listed as food products on their platforms. Regulators ordered the immediate removal of all datura-related listings and advertisements from these services. The action was taken due to serious public health concerns, as datura is a highly toxic plant not fit for consumption. Consuming unpurified datura can cause severe health consequences, including coma and death in extreme cases.

Ex-cricketers warn Jaiswal after ICC fine for clash with Fernando in 2nd Test

Yashasvi Jaiswal was involved in a heated altercation with Sri Lanka's Asitha Fernando during the second Test match following his dismissal. The ICC penalised both players with fines and demerit points in response to the incident. Former Indian cricketers Pujara and Raman expressed disappointment, pointing to Jaiswal's frustration as a key factor behind the confrontation. Karim stressed that senior players carry a responsibility to model good conduct for young fans of the sport.