pledess

His top students were capable of entirely understanding Scheme within a day or so (but not capable of entirely understanding all of Python and all of PyPI). He wanted students to be even better than that. He wanted them to lead productive and resilient collaborations even when they didn't or couldn't entirely understand the small parts.

Reformulating this slightly:

Yesterday, you saw the first 30-minute segment. Today, you arrive at the multiplex, and are informed that the three 30-minute segments (1, 2, 3) are starting shortly on screens A, B, and C. However, you are not told the mapping of segment to screen. You are asked to select a screen, and choose screen B. You are then informed about the status of either screen A or screen C: that status may be that it will play segment 1 (from your perspective: a duplicate) or that it will play segment 3 (from your perspective: out of order). Finally, you are asked whether you will be watching screen B, or the other, unrevealed screen. (If you don't actually watch your final choice, you're banned from the multiplex forever.)

Is this harder (e.g., not solvable at all) compared to the Monty Hall problem, because segment 1 is merely an annoyance, but segment 3 is a spoiler (permanently impacting your enjoyment of the movie)?

In my experience, things you can ask include:

What's the history of this first project I've been assigned to?

Who are the most important customers, or types of customers?

What's the risk tolerance (unless you've been told Move Fast Break Things)?

Things you usually can't ask directly, but need to learn quickly:

Is this a meritocracy? If not, what other factors matter?

What types of actions are perceived as throwing a co-worker under the bus?

Is most of my job to figure out what my job is (i.e., exploring how I can contribute most effectively)?

For "With your threat model in mind, they should identify opportunities to add new test cases," one common reason is that security engineers are shared across a large company and it may be very expensive for them to learn the different testing frameworks used on many different projects. Also, independent review (without any exposure to developers' conceptions about what should be tested, or why, or how) may be economically justified because outcomes of security bugs are sometimes much worse than outcomes of many categories of ordinary bugs. Other reasons may include that the security engineers want to run a test that can't be expressed in your testing framework without a huge change to the framework, they may want to develop their test cases adaptively such that most of the tests turn out to be useless and the cost of capturing every test under version contol may be very high, they may want to run tests from a commercial testing product for which the license does not allow bulk copying of the tests into a customer's testing framework, or (if they aren't in-house engineers) their business model is that they won't tell you every test that was run unless there's an associated defect finding.

We added ChatGPT Operator to UI testing, starting soon after it launched. It's only used as an extra testing step on top of everything we had previously used. A quick summary is: on the plus side, it sometimes gives us a much faster feedback cycle. On the minus side, it sometimes dives headfirst into advanced UI features, and can't find a way to backtrack when it makes a mistake there.

The letter mentions OAuth but doesn't mention the ongoing work to address the https://eprint.iacr.org/2025/629 findings, CVE-2025-27371.

Both rendering and security issues were relevant. Some of this is discussed under "Deprecate HTML Imports" at https://developer.chrome.com/blog/chrome-70-deps-rems

The TOCTOU is relevant (without suid) if someone can quickly make the right prediction of the tmpname2 value that's generated by the PRNG used by mkstemp, and create a symlink with that value before gunzip is executed. After calling mkstemp, the code should use the returned file descriptor, and thereby eliminate all TOCTOU risk. However, on (perhaps?) most devices that would realistically use atop, the PRNG works well enough that that prediction would fail.

https://newscience.org/nih/ suggests that the higher indirect rates at private research institutions may occur because "universities do subsidize research out of their own pockets."

There is a new comment by antirez in the past few minutes: https://github.com/redis-rs/redis-rs/issues/1419#issuecommen...

This may have unintended consequences on chances of a successful application. Now, as a high school senior, you have to compete against an additional pool of strong students who aren't especially interested in MIT's offerings, but have parents pushing them toward the least expensive of all top universities.

it may actually be reasonable to start a page under https://wiki.archiveteam.org/index.php/Category:Closing_proj... to track this

I thought "the Copilot coding assistant was trained on open source software hosted on GitHub and as such would suggest snippets from those public projects to other programmers without care for licenses" was explicitly allowed by the GitHub Terms of Service: https://docs.github.com/en/site-policy/github-terms/github-t... "If you set your pages and repositories to be viewed publicly, you grant each User of GitHub a nonexclusive, worldwide license to use, display, and perform Your Content through the GitHub Service." In other words, in addition to what's allowed by the LICENSE file in your repo, you are also separately licensing your code "to use ... through the GitHub Service" and this would (in my interpretation) include use by Copilot for training, and use by Copilot to deliver snippets to any other GitHub user.

I think "Block if :card_count_for_ip_address_daily: > 4" might translate to "As a customer using Stripe, obtain an IPv6 address from your mobile network operator. Do not share Wi-Fi with a group, such as by renting a house with four college buddies or by having more than two teenage children."

The essence of the problem is that there's no standard pathname for a personal directory that's guaranteed to be on local disk, even if $HOME isn't. Consequently, people have relied on /var/tmp/$USER for this. There are realistically affected users who can't change the new defaults.

Cleaning up /var/tmp on a timer is relevant to this academic environment (desktop-based research computing):

1. Each Debian machine is used by only one graduate student, but students do not have root access.

2. Today, /var/tmp is the only persistent local directory where the student has write access ($HOME is on a network filesystem backed up by the university).

3. Within the student population, there is strong institutional memory that /var/tmp isn't backed up by the university and isn't extremely robust (e.g., RAID), but also that nothing there is automatically deleted.

4. Students use /var/tmp for hundreds of Gb of data from simulations that take days or weeks. $HOME is too small and too slow for this.

5. In practice, less than 1% of students lose data through disk failure, accidents, etc.

6. A much larger fraction of students will lose data when sysadmins, who didn't get the memo about the /var/tmp change and thus haven't addressed the ingrained institutional memory, deploy new Debian machines.

7. Some of the students who lose data won't graduate on time.

For example, GET /tickets/12/messages/5 has these advantages over api_get_message_from_ticket?ticket_id=12&message_id=5

Avoids writing everything twice: you don't need to name the data fields both in the base URL and in the query string

If there are several parameters, writing everything twice may make the URL longer than one physical line in a text editor

The ? and & characters need to be quoted in most shells

The _ characters are sometimes hard to read if the entire URL is underlined

Names with api_json don't make it clear whether the request body must be sent as JSON, the response will be JSON, or both

Another name conflict is with https://github.com/package-url/purl-spec - which is used for software identification (e.g., see the https://www.cisa.gov/sites/default/files/2023-10/Software-Id... report).

with Gemini, maybe every video that says "comment down below" can be automatically modified to say "comment on the right"

It's not quite as bad as hunger-based pricing where EatUp would cost more on longer flights.

Many U.S. hotels changed that after the Mandalay Bay hotel incident in October 2017. A guest can no longer assume that their deadbolted hotel room door will only be opened in an emergency. Routinely, hotel staff (not accompanied by police) may knock and then immediately open a guest's door for what they consider a "welfare check" (e.g., guest has had a Do Not Disturb sign for 2 days). And, yes, guests may be strongly opposed to this for a variety of reasons (in the room but undressed, etc.) but it often is part of a hotel's normal operating practices. One of many references: https://www.reddit.com/r/askhotels/comments/vaxae2/comment/i...

I guess there's also a chance that a banking app will initially work but then fail after a forced update at a very inconvenient time. Is there a possibility that some use of "integrity signals" (in SafetyNet Attestation API and Play Integrity API) will be banned in the EU out of antitrust concerns? https://developer.android.com/privacy-and-security/safetynet...

I'm wondering about the probability that, out of all the affected customers, at least one had the research skills and social skills to identify another customer and successfully ask to meet. Like, for an essay about "His schnauzer needed a mom. WyZettle: the amazing story of a pivot from a home camera service to a dating app."

Almost nobody thinks of "earthbound" as suggesting a far-future flight from another world to earth. Instead, the connotations of earthbound are risk aversion, hesitancy to explore and discover, and fear of change. Many people are familiar with the phrase "earthbound misfit" from "Learning to Fly" (a Pink Floyd song). If you're trying to introduce an innovative and exciting product line, "earthbound" isn't it.

"lavabo" sounds too close to "lavatory" - and this has even stronger negative connotations. For Americans, at least, a "lavatory" (and thus perhaps the entire "lav-" word stem that means "wash") is never something that you have in your home. The term "lavatory" would most commonly be used at a school, but may also often be used at any other non-residential building. For Americans, "lavatory" almost always means a room that you visit to eliminate your bodily wastes (it doesn't only mean where you would wash afterward). Because anyone in the building can use the lavatory, it's usually not thought to be fully sanitary.

Potential customers will hate both of these, but "lavabo" is sure to kill U.S. sales.

Another possibility is ML for Cued Speech: https://arxiv.org/abs/2212.01083

Why is a Microsoft Azure team now using "Radius" to mean something different from what another Microsoft Azure team documented as recently as last week?

https://learn.microsoft.com/en-us/azure/active-directory/aut...

Zero subscriptions. Also, if a website prompts me to enter my email address for a newsletter signup, it's much less likely that I'll visit that website again.

I don't know anything about this specific applicant. Often, an applicant to a top school meets with a volunteer alumni/ae interviewer from the applicant's local area. They might get a competent interviewer with a dev background. There normally won't be a quiz on software development but there will be a conversation about a project. If the applicant asserts they did 95% of a project's work with 5% help from a parent or other contributor, but the applicant's responses give the interviewer a much different perception of the percentage, this can be flagged as an embellished claim. And that can sometimes be much, much worse than not mentioning the project at all.

It possibly matters because (at least in the U.S.) this specific data element has substantially different legal obligations than anything else that might be placed into a customer data structure: https://www.govinfo.gov/content/pkg/CFR-2012-title12-vol1/xm...

"(i) General rule. No national bank, and no director, officer, employee, or agent of a national bank, shall disclose a SAR or any information that would reveal the existence of a SAR. Any national bank, and any director, officer, employee, or agent of any national bank that is subpoenaed or otherwise requested to disclose a SAR, or any information that would reveal the existence of a SAR, shall decline to produce the SAR or such information, ..."

A bank might not want to aggregate data, within one IT system, if part of the data has the very unusual property that a subpoena must be declined.

How is data related to Bank Secrecy Act requirements (or similar requirements in a non-U.S. jurisdiction) stored? For example, a U.S. bank must contact regulators about a cash payment of more than $10,000. The payment itself is not information that must be kept secret from the customer, e.g., the bank can share information with the customer about the specific amount, date, etc. However, the act of reporting to regulators cannot be shared with the customer. The question is, from an IT perspective, is the act of reporting part of a "customer data structure"? Is it possible for IT staff to do simple database queries such as "date_that_customer_began < 2022-01-01 and large_cash_payments > 0"? Or is Bank Secrecy Act reporting data segregated into separate database tables or even separate IT systems?

I'd prefer a different solution in which the ecosystem of off-the-shelf consumer-grade routers and behind-the-router IoT devices cooperates to block Internet access, for a single IoT device, if that device is in a suitably exploitable state.

This isn't a perfect solution but your "support their devices with security updates for a reasonable amount of time" is a non-starter. For example, suppose I'm designing a doll for the Christmas 2024 season. The doll uses the Internet because it's an AI product that converses with young children about the latest STEM news. I don't know how long it'll be used: maybe my eight year old daughter will just find it boring, or maybe she'll physically destroy the doll because she disagrees with its opinion on the Riemann hypothesis.

I can't afford to maintain firmware beyond January 2025. If I have to commit, I'll just never release the product, and children will potentially have worse learning outcomes forever. But I am willing to have my 1.0 firmware send beacon frames to cooperating routers, announcing that my combination of product ID and patch level is a8217a61-09de-4b1e-8a99-b6fbc180cdce, and please blackhole me if this is a dangerous version. This requires more engineering to work effectively, but please don't stifle innovation by small IoT vendors who cannot commit firmware-maintenance resources to a product with an unknown revenue stream.