标签：#an

South Korean authorities issued the record-breaking fine following a data breach that affected over 30 million customers.

The general shape of the problem is that every public LLM benchmark is on a saturation clock that runs from the moment of its publication to the moment a model's training corpus has eaten it. The clock has been running, on the visible benchmarks of the last five years, for somewhere between twelve and thirty months before each one is no longer useful for differentiating frontier models. The benchmarks are not failing. They are doing exactly what they were designed to do, in the order they were designed to do it, and the field has been running through them faster than the people designing them anticipated. I want to put numbers on the saturation pattern, walk through what the contamination evidence actually says, and then sit with the question of what an honest benchmark would have to look like in 2026 — because the "private held-out eval" answer that the labs are converging on has economics that are worth examining carefully before any of us salute it as the solution. The saturation timeline, with numbers HumanEval (Chen et al., OpenAI, July 2021). 164 hand-written Python problems. The benchmark was published with Codex at 28.8% pass@1; the underlying GPT-3 base model scored 0%. GPT-4 (March 2023) hit 67% in the original Technical Report. By late 2024, OpenAI's o1-preview and o1-mini both reached 96.3% pass@1 ; Claude 3.5 Sonnet sat at 93.7%. The benchmark is saturated in the operational sense — the relative spread across the top ten models is around 10 percentage points, which is too small a gap to differentiate them on, and most of the new models arrive within a percentage point or two of the ceiling. The successor variants (HumanEval+ from EvalPlus, with augmented test cases) are the field's response. Lifespan from publication to operational saturation: about 36 months. MMLU (Hendrycks et al., September 2020). 57 subjects, ~14,000 multiple-choice questions, taken from publicly-available test prep and academic sources. The problem with MMLU is not that it's satura

The partnership will see TCS creating a business unit focused on deploying Anthropic's AI models to its customers.

Anthropic has apologized for stealthily throttling its new AI model, Claude Fable 5, with hidden guardrails that undermine both researchers and rivals using it to develop competing systems. The company says it is reversing course and will be more transparent about when the restrictions kick in, even if that means Fable refuses more queries. Fable […]

The surveillance company Leonardo wants more data : A surveillance company plans to add sensors to automatic license plate readers (ALPRs) that would mean the devices, as well as capture the license plate of passing vehicles, would also sweep up unique identifiers of mobile phones, wearables, and other Bluetooth-enabled devices in those cars, potentially letting law enforcement identify specific drivers or passengers. The technology, called SignalTrace, would turn ALPR cameras from devices focused on tracking cars to ones that can more readily track the location of particular people. ALPR cameras have become a commonly deployed technology all across the U.S.; SignalTrace would make some of those cameras capable of collecting much more data...

New research suggests that dozens of monetized YouTube channels are run by people and organizations that the US government has sanctioned for their ties to Tehran.

It’s the first estimate of how many Americans are sneaking onto Polymarket’s banned crypto-based platform.

This impressively light carbon commuter makes switching to an ebike easier than ever.

It might not feel all that different from older World Cups—for better or worse.

The games start June 11 and end with a grand finale in New Jersey on July 19. There are 104 of them. Here’s how to watch ’em all.

It’s a tale of two nuclear industries. In China, large reactors are coming together at a stunning pace. The country has nearly doubled its nuclear fleet since 2016, reaching nearly 60 gigawatts of total power capacity. The new facilities are nearly all gigawatt-scale pressurized-water reactors. Meanwhile, the US has built just two reactors in that…

In 2018, after nearly two decades working in Big Pharma, chemist Tim Cernak was ready to put his skills to a new use. For Merck, he’d developed precision therapies for cancer, HIV, and diabetes that could target disease while minimizing harm to healthy cells. But as a lifelong nature lover, he was increasingly concerned about…

PostgreSQL Partitioning for Multi-Tenant Audit Logs: Querying 100M Events Without Table Scans I'll be direct: if you're running a SaaS with compliance requirements and your audit_logs table is approaching 50M rows, you're three months away from pain. I've watched audit queries go from 200ms to 8 seconds in production at 2am because someone ran a "give me all logs for tenant X" report. Partitioning isn't optimization theater—it's table-stakes infrastructure. At CitizenApp, we store 9 months of audit logs across 50+ tenants. Without partitioning, a single compliance query would full-table scan 100M+ rows. With it, we hit the same data in <100ms. This post is exactly how we do it. Why Partitioning Matters (The Reality Check) Most developers treat audit_logs like any other table. You add an index on tenant_id and created_at , call it done, and move on. Then your compliance officer runs a query like: SELECT * FROM audit_logs WHERE tenant_id = 'acme-corp' AND created_at >= '2024-01-01' ORDER BY created_at DESC ; At 50M rows, even with a composite index, PostgreSQL has to: Index scan → finds millions of matching rows Random I/O all over the table Spill to disk if sorting is large Hope the OS cache is warm Partitioning solves this by eliminating the data you don't need from day one . Instead of scanning a 100GB table and filtering it down, PostgreSQL can skip entire partitions. A query against January 2024 data simply ignores partitions for February–December. I prefer partitioning because it's native PostgreSQL—no external caching layer, no read replicas, no Redis gymnastics. It's boring infrastructure that works. The Partitioning Strategy: Composite Partitioning (Range + List) I use a two-level partitioning scheme: Range partition by month ( created_at ) — keeps each partition to ~5–10GB List subpartition by tenant — ensures compliance queries are single-partition scans This is deliberately opinionated. You could do range-only, but then a multi-tenant query still scans the

OpenAI's GPT-5.5, GPT-5.4, and Codex are now generally available on Amazon Bedrock, one month after OpenAI revised its exclusive Azure arrangement. Pricing matches OpenAI's direct rates with usage counting toward AWS commitments. Codex shifts to pay-per-token billing with no seat fees. GPT-5.4 is the first OpenAI model available in AWS GovCloud. By Steef-Jan Wiggers

Cindy Zhang discusses the evolution of XDS, a unified UI system powering 10,000+ internal tools. She shares actionable insights for architects and engineering leaders on managing large-scale community contributions, executing safe monorepo refactors using JS AST and AI codemods, mitigating breaking changes via feature flags, and expanding UI libraries into full-stack platform systems. By Cindy Zhang

European economy airline Ryanair is under investigation in the UK for charging parents mandatory fees to sit with their children. The Competition and Markets Authority (CMA) said it was looking into whether the seating fees, which may be charging parents for the airline to meet its child safety and disability‑related obligations, are "unfair" under consumer […]

I spent yesterday building purejq , a pure-Python implementation of jq. I expected it to be the slow-but-portable option. Then I benchmarked it against the jq package on PyPI (the C bindings everyone uses to run jq from Python) and got this, on a 100k-object array, in-process: workload purejq jq PyPI (C bindings) field-access stream 9 ms 368 ms filter + count 55 ms 442 ms map + aggregate 18 ms 444 ms group_by 112 ms 704 ms transform + sort 136 ms 899 ms Pure Python, 7-40x faster than the C extension. That number looked wrong to me too, so before publishing anything I made the benchmark script verify every output against the actual jq binary first ( tools/bench.py --verify ), re-ran everything as median-of-7, and gave the bindings their best-case API. The gap is real. Here's why. The serialization tax The C bindings wrap real jq, and real jq only speaks JSON. So every call does this: your dicts -> JSON text -> C parser -> jq evaluates -> JSON text -> dicts That round trip costs about 350-450 ms for 100k small objects on my machine, before any actual filtering happens. You can see it in the numbers: even a trivial field access pays the same ~400 ms floor as a group_by. purejq skips the trip entirely. It compiles the jq program once into Python closures and walks your dicts and lists directly: import purejq prog = purejq . compile ( " group_by(.team) | map({team: .[0].team, n: length}) " ) prog . first ( data ) # operates on your objects, no serialization The lesson generalizes beyond jq: when you embed a C library that has its own data model, the marshaling boundary is often more expensive than the work. An interpreter written in your language gets to skip the boundary, and that can buy back an order of magnitude. Surprise number two: the CLI beats the jq binary on big files This one I really didn't expect. End to end on a 93 MB file (1M objects), parse + filter + output: workload purejq CLI jq 1.8.1 binary single lookup 0.51 s 1.68 s filter + count 1.08 s 1.96 s grou

If you doubted his genius, doubt no more.

I built a knowledge graph out of my own work sessions. Hundreds of them — transcripts of me building a system with LLMs, extracted into concepts, decisions, findings, and the edges between them. For a while it felt like the thing was working. I'd query it, get back a clean structured answer, and move on. Then I ran a foreign model against it. I gave a different model my concept definitions and asked it to reconstruct the system, both the vocabulary and the relationships. It recovered 97.7% of the words. It recovered 61.1% of the structure. That 36-point gap was the first time I could see the problem instead of just living inside it. The vocabulary transferred because the definitions were written carefully. The edges didn't, because the edges were the part I'd let the extraction handle. And the whole time, querying the graph had felt complete. The structure came back typed, connected, confident-looking — so I stopped looking. I started calling it premature retrieval closure: the retrieval returns something shaped like a whole answer, which is exactly why I didn't notice the parts that were missing. Part 10 of Building at the Edges of LLM Tooling . If you're running a long-term project through an LLM-backed memory system (anything that turns raw sessions into structured, persistent memory), this is about the step where the structure starts lying about how complete it is. Start here . Why It Breaks Every memory system of this kind does the same move. An LLM reads raw interaction (a conversation, a document, a session log) and lifts structured memory out of it: entities, facts, rules, summaries. That structured memory becomes the thing the agent reads later, instead of the raw record. The lift is where fidelity goes. Pulling clean structure out of messy text means making decisions the text didn't make explicit: which entity this pronoun refers to, whether a relationship is real or inferred, what to keep and what to drop. Those decisions can be wrong, and when they are,

The company changed course after researchers spoke out against the policy, which would have covertly limited Claude’s ability to develop competing AI models.