Writing, page 4

AI Agents Finished 167 of 400 Psychology Experiments. Their Data Looked Human on the Biases, Not the Abilities.

7 minute read

· Agents

CogArena sends AI agents through ten online psychology experiments in a real browser. Four economy-tier models finished 167 of 400 sessions; most failures were the agent declaring itself done early. Where the data could be scored, agents reproduced human bias signatures 66% of the time and human capacity signatures 31% of the time, which matters for anyone worried about agents slipping into online studies.

One Crawlable Page in Four Has AI Text. At the Center of the Web It Is One in Nine.

7 minute read

· Benchmarks & measurement

A new study of more than 5 million crawlable pages estimates that about 24% carried AI-generated or AI-edited text in August 2026. The rate is 11.6% in the web’s most central tier and 27.5% at its periphery, which holds most of the pages and nearly four-fifths of the growth. It is about 6% on .gov and .edu, and twice as high in English as in other languages.

The Rejection Came Back 200

17 minute read

· Security & privacy

For thirteen days, any prompt to GLM-5.3 on Cloudflare Workers AI longer than 681,280 tokens could come back with pieces of other customers’ requests in it. Cloudflare counts 155 such responses. The engine had rejected those prompts and then served them anyway, with an HTTP 200 and a coherent answer. The same bug had been filed upstream in July as a source of garbage output. What turned garbage into a leak was a full cache, and the only test standing in the way checked the status code.

The Score Survived Compaction. About One Rule in Three Did Not.

19 minute read

· Agents

A TΓΌbingen group ran four frontier models through OpenCode on three long-horizon benchmarks and found that when the agent compacts its context barely matters to the score. The same summaries drop about one of three anti-cheating rules the first time they are written and about two by the tenth, and on real coding sessions they lose instructions about how to work six times as often as instructions about what to build. The benchmark can’t see that, because everything the score needs is on disk.

The Old Network Names More Lesions. The Question Is Which Ones It Sends Home.

10 minute read

· Science & medicine

A dermatologist tested a 2015 image network and two frontier language models on 200 biopsied skin lesions. The network named the right diagnosis far more often. On the decision a patient cares about, whether a lesion can be sent home, the comparison only means something once the two are set to the same level of caution, and at its default setting the network discharged 18 of 30 melanomas.

Harder to Read Than People, and Easier Than Last Time on 19 of 21 Counts

7 minute read

· Benchmarks & measurement

A new paper finds that frontier models write with more of the features that slow human readers than people do. Its starting point is a sense that AI text has recently become harder to read, but its measurements compare models with people at one moment. The one before-and-after pair it reports, Claude Opus 5 to Opus 5.5, moved toward easier text on 19 of 21 comparisons.

A 75% Price Cut That Leaves the Chat Bill Unchanged

7 minute read

· Models, inference & pricing

Frontier API prices have fallen by up to 80% since June, Max Woolf writes. The headline cuts land on different parts of the bill. Fable 5.1’s cache reads are 75% cheaper while its input and output prices did not move, so a chat workload pays the same as before. On an agent workload I worked through, Opus 5.5 is about 28% cheaper than Opus 5, not 40 to 60.

Twice, Four Times or 1.7 Times: The Multiple Depends on Which Doctors

9 minute read

· Science & medicine

Microsoft’s diagnostic orchestrator is now in Science, and the coverage says it is twice as good as doctors. The same post also says nearly four times and 1.7 times. All three are true, and each is a different comparison: the physicians had references or did not, and the cases were NEJM puzzles or ordinary emergency admissions.

239 Proved, 60 in Public, and the Statement Was a Fuzz Test

22 minute read

· Security & privacy

Anthropic and Theorem say Claude proved 239 OSS-Fuzz entry points across 123 libraries free of crashes, on compiled binaries, with a proof checker as the referee. The public repository holds proofs for 60 entry points in 25 libraries, and 138 proved statements out of 20,164 stated. The theorems are real. What they cover was decided years ago by whoever wrote each fuzz test, and the binary they are about is a static RISC-V build, not the x86-64 builds OSS-Fuzz runs or the dynamic glibc builds most people deploy.

Five Finders, One Fix, and a Honeypot Hit on the 17th

32 minute read

· Security & privacy

The newsletter headline said AI found a security hole in 70% of the web. The PHP Foundation’s own account says a contractor found the first signs in 2024 while fixing an audit finding, three firms reported the full impact in August and September, and the release came 71 days after the report that made it urgent. Chainguard’s chief executive says a researcher told him on Friday 16 October that something was coming in PHP and that he found the bugs himself that night. Amazon’s honeypot logged an exploitation attempt the next day, three days before the first leak, and OpenAI says frontier AI found the flaws before attackers did.

Zero Above the Line, Some Below It

26 minute read

· AI safety & governance

Less than two weeks apart, Sam Altman told the UN Security Council that no level of catastrophic risk anyone would name, down to a tenth of a percent, is acceptable, and told a podcast that the world should accept some bad things happening, and that he would not promise there would be no major hacks, misuse or scams. Read together they are one position, and it is the position OpenAI’s written policies already take. What none of the documents say is how much of the accepted kind is too much.

Twelve Steps, Coarse to Fine

7 minute read

· Models, inference & pricing

Hazy Research’s Next-Concept Modeling generates text in a fixed hierarchy of short-to-long code sequences instead of 128 denoising passes over the whole thing. The paper reports 10 to 31 times faster generation than diffusion baselines. The speedup is real in the tables, and the quality claims need reading one row at a time.

Same Weights, Different Harness, Up to 11.5 Points

10 minute read

· Agents

Google Research has released HarnessΒ², an Apache 2.0 framework that lets an agent rewrite its own prompts, tools, memory and skills while the model stays frozen. The paper’s gains are real and largest where the starting harness was weakest, and the judging setup deserves a close read.

Six to One, on a Clock That Counts the Waiting

24 minute read

· Business & work

Anthropic’s new employment report says its agents worked six hours for every human hour by August. The clock behind that number keeps every pause under half an hour, and a second clock in the same report finds sessions about 80 percent idle. Between the two, the active-work ratio could be anywhere from 1.2 to 6. The junior-hiring figures have the same problem in reverse: every share fell, and the count most likely rose.

page 4 of 12 · 178 essays