sequin

This is never going to be resolved through speculation. There are parallels with the Zodiac killer's identity where there's a group of very likely candidates but no conclusive proof to definitively settle the question one way or the other. Outright admissions mean nothing (Wright). Even cryptographic proof could leave some margin of doubt (speculations of an ECDLP break or key theft).

Resist hypes and just use whatever you feel like. Torvalds uses a 40 year old EMACS implementation and that seems to be working for him.

I left Spotify years ago. Youtube is so much nicer in terms of content alone. But Youtube, with its insane backlog of video's not available elsewhere, is straight up a monopoly, so they too will start squeezing customers at some point. In anticipation of that I've been collecting flacs again. It's actually kind of a nice hobby.

I'm just so glad I'm past the phase where I'd worry about these things. It's very exhausting.

I doubt talking heads on the TV can move markets.

How can the models be impressive if they switch to Chinese mid-sentence? I've observed those bizarre bugs too. Even GPT-3 didn't have those. Maybe GPT-2 did. It's actually impressive that they managed to botch it so badly.

Google is great at some things, but this isn't it.

The Weezer video was quirky, funny, creative, catchy and appealing to multiple generations. A stroke of genius by Microsoft.

Probably a prank or an ambush set up by the anti-billionaires.

FWIW, I just gave Deepseek the same prompt and it solved it too (much faster than the 41m of ChatGPT). I then gave both proofs to Opus and it confirmed their equivalence.

The answer is yes. Assume, for the sake of contradiction, that there exists an \(\epsilon > 0\) such that for every \(k\), there exists a choice of congruence classes \(a_1^{(k)}, \dots, a_k^{(k)}\) for which the set of integers not covered by the first \(k\) congruences has density at least \(\epsilon\).

For each \(k\), let \(F_k\) be the set of all infinite sequences of residues \((a_i)_{i=1}^\infty\) such that the uncovered set from the first \(k\) congruences has density at least \(\epsilon\). Each \(F_k\) is nonempty (by assumption) and closed in the product topology (since it depends only on the first \(k\) coordinates). Moreover, \(F_{k+1} \subseteq F_k\) because adding a congruence can only reduce the uncovered set. By the compactness of the product of finite sets, \(\bigcap_{k \ge 1} F_k\) is nonempty.

Choose an infinite sequence \((a_i) \in \bigcap_{k \ge 1} F_k\). For this sequence, let \(U_k\) be the set of integers not covered by the first \(k\) congruences, and let \(d_k\) be the density of \(U_k\). Then \(d_k \ge \epsilon\) for all \(k\). Since \(U_{k+1} \subseteq U_k\), the sets \(U_k\) are decreasing and periodic, and their intersection \(U = \bigcap_{k \ge 1} U_k\) has density \(d = \lim_{k \to \infty} d_k \ge \epsilon\). However, by hypothesis, for any choice of residues, the uncovered set has density \(0\), a contradiction.

Therefore, for every \(\epsilon > 0\), there exists a \(k\) such that for every choice of congruence classes \(a_i\), the density of integers not covered by the first \(k\) congruences is less than \(\epsilon\).

\boxed{\text{Yes}}

Rather the opposite. A vibe-coded startup cannot survive if it can be trivially duplicated. The proof will be in observing the inverse phenomenon; (pure) software companies disappearing.

It's certainly morally and legally dubious to facilitate attacks on things that others choose to use in within their own private domains, just because you disagree with that choice. But that's how these people roll.

I honestly don't understand why a terminal emulator needs to be performant. Seems like peak bikeshedding to me.

Very, very few people care about privacy to a meaningful degree, even if they (loudly) profess that they do. For communication apps specifically, there's also the network effect problem and having to deal with the increasing proliferation of government speech laws (Online Safety Act and what have you).

If you just want to make a buck, build a ChatGPT wrapper where people will pay you for the privilege of uploading their deepest secrets and intellectual property to your servers.

If you're ideologically motivated, forget about the profit motive and go FOSS.

Many people (not the EFF or ACLU, lol) saw it coming years ago and put in a monumental effort to try to turn to tide, only to be relentlessly repressed and dehumanized by bolsheviks and idiots.

I'm the same. For my use case there is little benefit in using those tools, and for security and privacy reasons I don't want to unleash all this software on my computers and network.

Same; any time I see BWT I think of Fabian Giesen. I think he was only 16 or 17 when he wrote that article.

By chance I once sat near him and the rest of Farbrausch at a demoparty, but I was too shy to say hi.

What a miserable and hopeless undertaking having to do this tightrope act of appeasing existing users, prospective users, advertisers and lawmakers. Managing a fledgling social media platform emerging out of ideological fault lines while others rake it in with GPT wrappers is an exercise in masochism.

How did they get the passwords to his Google and Coinbase accounts? He reused passwords? The same one for Google as for Coinbase? Or did they reset his Coinbase password via his Gmail? The post doesn't make this explicit, but it warns against password reuse.

These are all very circumstantial.

Considering the need for LLM boilerplate coding as a language and tooling failure is an interesting observation. It begs the question why there is still such a thing as boilerplate. I mean I understand how programming languages work and why boilerplate is essential, but at the same time LLMs demonstrate that brevity is possible without loss of expressivity or legibility, albeit in a very roundabout and costly way. Perhaps there is a programming paradigm that strikes a balance between these opposites, especially if you're willing to spend more computational resources on it than traditional compilation (but with a complexity ceiling orders of magnitude lower than LLMs).

This could have been an amazing short sci-fi story.

Making sensible choices requires thinking things through, and understanding what it really is that you are doing, or refrain from doing it. Experts are just as prone to neglecting this as novices. But expressing your misgivings about prevailing orthodoxies is an ungrateful undertaking as it tends to make people angry, and there is little honour to be had in getting vindicated.

Around $15/month for Claude (yearly subscription they offered me at the start of the year).

Had a Gemini Pro sub for 1 or 2 months but it didn't impress me that much.

When I need real horsepower (usually advanced math stuff) I use DeepSeek, which is both free and unparalleled in my opinion.

I did some experiments with OpenRouter but my total usage is still below $10.

More and more I've been experimenting with local thinking models and while it's quite slow (on my non-Apple, CPU-heavy hardware, at least), for some use cases it's an acceptable trade-off and the results have been satisfactory.

I care a lot about data privacy and I'm just not gonna upload clients' proprietary codebases under NDA to some AI company. I've been considering buying or renting proper hardware for local inference but the TCO (particularly with regards to depreciation) is not intuitive.

I don't get this. I have no education, no resume, no Linkedin and I've never applied for a job in my life. Employers/recruiters reach out to me, and I usually turn them down. I'm not humblebragging, I'm not better at my job than most people. What seemed to have worked for me was doing tons of FOSS work (unpaid, in my spare time). People started noticing me and it naturally picked up from there. I'd wager that actual, observable software engineering (FOSS) is simply a closer proxy for job performance than accreditation and past experience.

In this day and age, text is abundantly accessible at all levels of complexity, and you can generate infinitely more with LLMs. Affordability is not an obstacle to developing reading comprehension, and hasn't been for a long time.

GitHub is a completely different scale. If Google hadn't wasted its opportunities to become the prime source code platform they'd also have issues.

High school dropout, never attended college. I never had any problems getting work. Having done extensive open source work has helped a lot. YMMV.

It seems that at least in software engineering, Claude is popular and people are shoveling tons of money into it, whereas non-tech people are not inclined to pay for ChatGPT. Market share might not be so important if your competitor is raking in all the cash.

She :)