> there is no way to make the distinction in an interview
Just ask?
Some online degrees state that they're equivalent, but interviewers may still have their own opinions. I would discourage anyone from failing to mention the online nature of a degree in their CV. You're really not doing yourself a favor. A rigorous online degree is something to be proud of. I see people with PhD's proudly announcing their online course certificates on LinkedIn. However, 'discovering' that an education was of a different nature than one had assumed based on the presented materials may raise questions.
The article says that having decompiled some functions helps with decompiling others, so it seems like more than one example could be provided in the context. I think the OP was referring to the fact that only a single prompt created by a human was used. But then it goes off into what appears to be an agentic loop with no hard stopping conditions outside of what the agent decides.
We're essentially trying to map 'traditional' ML terminology to LLMs, it's natural that it'll take some time to get settled. I just thought that one-shot isn't an ideal name for something that might go off into an arbitrarily long loop.
> The ‘give up after ten attempts’ threshold aims to prevent Claude from wasting tokens when further progress is unlikely. It was only partially successful, as Claude would still sometimes make dozens of attempts.
Not what I would have expected from a 'one-shot'. Maybe self-supervised would be a more suitable term?
Sounds like saying that nothing about the Industrial Revolution was steam-machine-specific. Cost changes can still represent fundamental shifts in terms of what's possible, "cost" here is just an economists' way of saying technology.
Great initiative and beautiful site! Tiny nitpick, the wrapping of the controls above the table on my phone could probably be improved. What did you use for the table?
There's a school of thought of using mostly base R, for all its flaws it already had before Hadley, and selectively using some tidyverse packages. Base R has been the de-facto coding standard for academic statisticians for decades, with all the wealth of open source packages that that entails, and some of the tidyverse packages are just a godsend. ggplot2 is probably the most powerful plotting library I've seen, while being fairly accessible. You don't have to subscribe to an entire philosophy for data wrangling or plotting (and may even frown at the syntax overloading) to get a huge amount of utility out of it.
The article mentions some complications and browsers like LibreWolf that address some of them. But that's not really what I have in mind - I'm thinking more of a dashboard with editable fields for everything that might get queried, populated with what currently gets sent, and a drop-down that lets you select among preset/saved custom profiles.
What I don't get, all this data is reported by your machine - why isn't there a tool/browser fork that allows spoofing a (fairly) complete realistic profile, with some sane presets like Edge/W11/Thinkpad or Safari/macOS/M4? Is it too complex, would it break too much, or am I just unaware?
Fully agree that text-based CAD is the obvious path forward. But OpenSCAD won't cut it, it just lacks too many features, starting from basic fillets to more fundamental things like relative object positioning. Check out CadQuery, it's much more ergonomic and future-proof.
Maybe Firefox would have a higher market share if they worked on features their users actually want instead of things that get widely criticized. I personally would use it a lot more if it had an --app flag like Chromium, which would probably also be a lot less work than AI integration.
Much of this thread is people complaining about on-by-default features and hard-to-find configurations. We're really just talking about a continuation of a trend.
> Honestly people who deny any usefulness of AI are getting dangerously close to flat-earthers by now.
What made you jump to that conclusion? My guess about someone who's using a non-mainstream browser and figuring out how to configure it to their liking is that they're likely also using AI in more ways than the standard Chat-Webinterface, eg Agents, CLI tools, MCP,... To give an analogy, rolling your eyes over brainrot memes isn't denying the usefulness of smartphones or messengers either. The underlying sentiment is being critical of things that get pushed down our throats through A/B-optimized patterns that ultimately serve other interests than your own, profits or darker.
In principle, you don't even need a "server", ie purpose-built hardware like what he's showing. An old laptop will be fine for many projects, although I'd be interested if someone can help put a perspective on how to decide whether your specific project/load will be too much for your specific hardware. My bigger concern with 'true' self-hosting would be how to get a good/reliable enough internet connection.
How is "most of that pressure gone"? You're still being evaluated and have to code against a clock, with less time, less opportunities to ask questions and less immediate feedback that could get you back on track.
Also, your 20 vs 30 minute calculation ignores that companies are incentivized to conduct more screening tests if it becomes practically free for them. But the number of positions stays the same. So if instead of 10 screening calls they do 16 tests for one position, that's already more time being wasted, even if the tests are 1/3 shorter. And realistically, the number will shoot up much more.
Do you realize that your product will only lead to more time being wasted on the side of the applicants, who are already the weaker party? How do you justify that?
Copyparty can't (and doesn't want to) replace Nextcloud for many use cases because it supports one-way sync only. The readme is pretty clear about that. I'm toying with the idea of combining it with Syncthing (for all those devices where I don't want to do a full sync), does anybody have experience with that? I've seen some posts that it can lead to extreme CPU usage when combined with other tools that read/write/index the same folders, but nothing specifically about Syncthing.
> verified loader apk, which in turn dynamically loads any apk the user wants
Wasn't this kind of solution considered and sort of dismissed (because of too much centralization iirc) by F-Droid (can't find the reference now)? It seems like something that's worth trying, but in the end it's just a band-aid. If it gets any traction Google will shut it down. The real disease is dependence on a duopoly of (quasi)-proprietary OS for the dominant computing platform of our time.
They're working hard on shutting that down as well with Passkeys. It's only a matter of time until the only way to log in will be through de-facto proprietary apps.
Stupid question: couldn't we work around that with some VM/container-style solution? They could probably find ways to lock it down with TPM/TEE and similar, but in today's landscape it should be possible if you're willing to accept the performance and battery cost. And if it does get traction, there'll also be more push to keep open alternatives viable. Giving in without a fight is the only way to ensure you'll lose.
Coming after the Austrian military ditched MS Office (365, Copilot, whatever it's called now) for LibreOffice [0]. Similar stuff going on in Denmark and to some extent in Germany [1]. Way to go!
> I personally know HRs who consider the "PhD" tag in a CV a red flag.
Which should only tell us what a cancer HR departments are. Filtering someone from talking to the hiring department, for the crime of doing a PhD, and telling others about it - to me this sounds more likely like someone getting a kick out of wielding the bit of power they have over others than a useful contribution to society.
> 90% of the reason I use chat gpt is that it doesn't have ads. Shove them in and usage goes way down.
Don't you think that people would have felt similarly about early Google search?
> any type of automation (the big promise of AI) ads don't matter
I would have thought that ads have no place in an OS, but they proved me wrong on that. How sure are you they won't prove you wrong on automation ("summarize this text for me - here you go, shall I run it through Grammarly now?", "build this app for me - [uses freemium tool instead of equally capable FOSS]",...)? Never underestimate the potential for enshittification.
I don't get why ads are never mentioned in the article. The current use cases of GenAI (chatbots, generate [art] for me,...) have extremely obvious monetization angles through ads, and then there's a positive chance that they can bring in revenue through more ways than that (they already do in the form of subscriptions, eg). It might be that the economics still don't work out, but at least it should be considered?
In case that wasn't clear, I didn't mean that every government in the world, nor even every agency in the same country, should be using the same office suite (which they pretty much do now). (Part of) the idea is to increase software diversity, for the same reasons that monopolies are generally bad in other areas. It's pretty clear that that goal alone will require switching to open document formats.