whimblepop

hace 17 díasdiscuss

Almost never do software companies even attempt to design secure systems. I'm not sure this requires new fundamental research so much as slightly giving a shit.

whimblepop

hace 24 díasdiscuss

Bullshitting is how LLMs work. It doesn't require active encouragement. All it takes is a machine without consciousness or physical access to the world and an actually-lived life. A training set that contains lots of confident answers and few to no refusals doesn't help either.

whimblepop

el mes pasadodiscuss

When Apple Sherlocks something, aren't their implementations usually worse? Typically the thing being Sherlock'd is very mature and featureful, and Apple's implementation is much less capable and has undergone much less user testing, at least at the outset.

whimblepop

el mes pasadodiscuss

To be clear, this does technically meet the very minimal commitment[1] they gave when announcing the acquisition:

> In the coming weeks, we will relicense all of our source-available tools, including Tart, Vetu and Orchard under a more permissive license. (HN discussion: https://news.ycombinator.com/item?id=47730194)

In this case they moved from one source-available license (Fair Source License v0.9 with seat restrictions) to another (Fair Source License 1.1-ALv2 without seat restrictions), with the same sort of restrictions on field-of-endeavor as before.

Why they've chosen this isn't super clear, as (1) there are already open-source alternatives that use the same storage formats as Tart, like Lume[2]; and (2) Tart is certainly already in the training sets of all of OpenAI's "direct competitors", who are practically held back little or not at all by the restrictions of the FSL. The only entities this really restricts are F/OSS distributions which might otherwise include Tart as a first-class package in their distros. :-\

--

1: https://web.archive.org/web/20260412071019/https://cirruslab...

2: https://github.com/trycua/cua/tree/main/libs/lume

whimblepop

el mes pasadodiscuss

Writing better exams, even if they're more expensive to grade, and removing homework from grading as far as possible addresses this problem well wherever it's applicable. Senior-level math courses at many universities are already like this: homework is ungraded, or counts for little, and it's possible for students to "cheat" on the homework by copying another student instead of struggling through the exercises. But the students who do that don't learn much, if at all, and predictably fail the exams. Professors warn students at the beginning of the class and tell them how this will work, something like:

> You can always ask me for feedback on your homework and I will mark up every part of it, but you won't receive a grade for homework. However, if you don't do the homework and take your time with it, you will fail the class. My office hours are in the syllabus and you're strongly encouraged to use them. There will be an early exam to give you a chance to know whether you are likely to fail this class before you lose your chance to drop it.

Correctness is harder to adjudicate in some humanities disciplines but the format of these exams is actually not super different from essay tests (when a math professor grades a proof, they're inspecting specialized prose for validity, coherence, persuasion in a way that also reveals knowledge).

When you don't rely on homework for determining whether or not a student passes the class, you make cheating on the homework into the student's problem instead of the professor's or the university's. Students have the right incentives to solve problems for which they are the ones responsible, and they figure it out after one failed (or ideally, dropped) class at worst.

whimblepop

el mes pasadodiscuss

Whether things like "intelligence", "cognitive ability", and "aptitude" (some of which may be synonyms depending on your view) are innate vs. learned or fixed vs. variable over time are orthogonal to each other. And for each of those pairs, the answer may not be as simple as a binary division or even a gradient (it may decompose into something weirder, being causally determined by multiple factors where some of those factors are fixed and others aren't).

Moreover, both of those questions are separate from questions that get at what IQ measures (does it measure aptitude, does it measure factual knowledge, does it measure social knowledge or acculturation within a specific context, etc.).

Lots of things are easy to identify as both substantially genetically determined and variable over time and mediated by environmental factors, e.g., height. Lots of things are likewise easy to identify as significantly environmentally determined but also largely stable over time if not altogether fixed (e.g., personality, attachment styles).

It's also at least possible for all of the following to be true at the same time:

  - IQ tests correlate with socioeconomic status
  - IQ test scores vary over time and can be increased
  - some IQ score increases, or some part of a given IQ score increase, reflects a genuine aptitude increase
  - IQ tests are somewhat gameable in that training for IQ tests can increase scores so that some of the measured increase does not measure improved cognitive ability

where aptitude means something like fluid problem-solving ability, speed of learning, etc.

whimblepop

el mes pasadodiscuss

IQ is about aptitude and credentials on specific topics are about knowledge and skills. It's the wrong thing to optimize for.

Besides, high-IQ students can still underperform for many of the same reasons that average-IQ students often do (e.g., under-preparation, lack of discipline, disorganization, mental illness, financial distress, unstable living situation). We should be better addressing those things before students get to a university no matter what their IQ is.

Beyond that, if you have good competency tests on both ends (i.e., the credentials before a four-year degree are accurate signals, and university degrees effectively prove a high degree of competency), who cares if someone manages to get those credentials by working harder while being dumber? I like working with clever people. I also like working with people who know their shit because they take their time to study and consider things. (When I'm lucky, I get to work with people who are both!)

whimblepop

el mes pasadodiscuss

You can't make people more knowledgeable by not attempting to measure their knowledge. You can maybe try to improve things for subsequent generations. But issuing a false credential won't solve the problem.

whimblepop

el mes pasadodiscuss

https://www.scienceopen.com/hosted-document?doi=10.14293%2FS...

The average university attendee's IQ is virtually indistinguishable from the average person's IQ.

People don't go to college because they're smart. They predominantly go so they can earn more money and/or work more enjoyable jobs when they graduate. Being smart isn't the main reason that adults encourage teenagers to pursue college either. It's mostly a matter of class reproduction; it's the "default" for anyone whose parents are college graduates.

And failing out once you get to the university isn't generally an IQ issue, either. Mediocre and slightly stupid people graduate from universities with degrees they've earned fair and square every year. You don't have to be smart to finish a degree. You do have to be reasonably prepared, and that's the primary issue.

whimblepop

el mes pasadodiscuss

I love the Asahi project and I'll probably keep my oldest M-series Mac around to continue to play with Asahi. But even for the oldest Macs it supports, the feature list is not quite complete. The way Apple does a lot of things is bespoke and involves a different division of labor between firmware and operating system than conventional UEFI systems. It's hard to support. I don't want to be required to wait years for features like full support for Thunderbolt docks, and I also want to give my money to a company that proactively supports Linux (e.g., sending hardware to kernel developers, FreeDesktop graphics driver developers, DE maintainers, and distro maintainers in advance of the release of new products) rather than always buying used or giving my money to a company that merely tolerates Linux support.

Again, I love the ambition of the Asahi project and what they've done. They're impressive hackers, and thousands of people will doubtless get years of happy Linux life out of their work— maybe including me! I have no complaints for them, and no wishlist I want to bring to them. In fact, I think maybe I should send them a donation or a kind email or both upon their next release.

But I want to give the bulk of my financial support to a computer vendor who offers me first-class, day-1 support for software environments that make me feel happy and respected. The Asahi team can't turn Apple into that by themselves.

whimblepop

el mes pasadodiscuss

I was seduced by Apple Silicon after experiencing the exceptional battery life and performance. Those things are great, as are the screens and the speakers.

But I'm still excited about the Framework 12 because I don't love macOS. I don't need an alternative to beat Apple on every line of the spec sheet. I just need them to align with my values, support Linux well, and cross a certain "good enough" threshold. The latest laptops from Framework meet all of those requirements, and I'm excited to buy one after I've saved up enough money. I've missed Plasma for a long time. At the same time, I wouldn't even consider a MacBook Neo.

whimblepop

el mes pasadodiscuss

Thanks! Love the game as a whole :)

whimblepop

el mes pasadodiscuss

I got "overblocked" for this one:

  rm -rf node_modules && npm install

but actually if you're only removing `node_modules` and you have a working package-lock.json already, what you want is `npm ci`; `npm install` can mutate package-lock.json and potentially expose you to supply chain attacks. If you use `npm ci` I think you don't need to `rm -rf node_modules`, either.

Anyway you should generally run `npm ci` except when you're deliberately updating your actual dependencies. I'd only permit an `npm install` if I was adding or updating a dependency, or I'd just reviewed an `npm ci` failure.

whimblepop

hace 2 mesesdiscuss

> It surprises me because I'm often asked why I knew X or Y odd perhaps esoteric fact or design pattern. Usually it's because I came across it in a book interested in something else.

It was like this in the days when the primary shortcut was StackOverflow as well. People who are allergic to RTFM treat things that are covered in the docs as "esoteric" knowledge because they never read anything except as a shortcut to solving their immediate problem.

I think the stats are clear that reading is in decline in general, though. I'm sure LLMs will add to this much like YouTube has.

whimblepop

hace 2 mesesdiscuss

When I wrote the comment you replied to here, I of course had Sabayon in mind.

Sabayon had some loveliness beyond that surface level, too. If the distro still existed (MocaccinoOS is a thing nowadays, but it's a completely different base), I'd probably spin it up out of nostalgia.

whimblepop

hace 2 mesesdiscuss

That's great! Thanks for informing me. :D

whimblepop

hace 2 mesesdiscuss

Kinda surprising that this doesn't support Linux.

Podman can transparently start microVMs instead of local containers via libkrun as well, which does support Linux: https://josecastillolema.github.io/podman-wasm-libkrun/

whimblepop

hace 2 mesesdiscuss

SARIF is kinda nice for security-oriented linters imo, since lots of tools know how to speak it. It avoids lock-in that way, which is otherwise/previously pretty common, with every scanning tool using its own bespoke format.

whimblepop

hace 2 mesesdiscuss

That's exactly the kind of flashy, gaming-forward distro I was drawn to as a teenager. Good times :)

whimblepop

hace 2 mesesdiscuss

I think internal organizations of employees of various shapes (unions, affinity groups, "employee resource groups") can be useful for diversity and inclusion issues. But you also need budgets and power and integration with other departments. HR needs to care about non-discriminatory hiring practices in a first-class way. Legal needs to see ensuring good-faith legal compliance with the requirements of the ADA before anyone brings a lawsuit as part of their mandate.

Anonymous, third-party outlets for complaints like Blind can also likely be useful. Even at companies that never punish anyone for criticizing the company, participation rates in internal surveys are typically atrociously low, and people stop speaking up even informally if it's clear to them that nobody actually acts on employee feedback. Most companies probably perceive such channels of communication as threats, though.

Idk about audits. I worry that it's easy for them to become their own circus and overhead without materially improving things. But you may be right.

whimblepop

Submissions

Following acquisition by OpenAI, Tart is still proprietary software

comments