yodon

Given all the recent press about bun rewriting 500k lines of zig into rust at a cost of $165K tokens of Fable in an 11 day sprint, would be super interested to hear about your experience and approach.

How is this not a the mother of all GDPR violations?

The advantage of their dataset is it's large enough that you have a statistical number of PR's, allowing you to treat the average cost of Python PR's vs the average cost of C# costs (or etc.) as statistically meaningful.

I wish they'd do a follow-on post drilling into the impact of the programming language on cost-per-task, specifically looking at cost to complete tasks in mainstream strongly typed languages (eg. C#, TypeScript) vs dynamic languages (eg. Python, JavaScript). Does the additional verbosity of the language help or hurt cost per task?

I was on my phone, so all I saw was a generic Safari error message. I think it was site not found, but I don't recall. I'm able to see it now, but am coming in through a different network provider in case that matters.

Also, not to be an apologist, but I trust you realize that "the way they handle US sanctions by blocking everybody from entire countries" is how US companies are required by law to act. Not all do act that way, of course, but almost any that have enough overseas customers to matter do, because they have enough overseas customers to need to be aware of those laws.

The code appears to be unavailable. This includes not just wordgard but all the ProseMirror code as well.

If the motivation for moving off GitHub was "GH is down too much", it might be worth tracking how many 9's of uptime is lost in the self-hosted case.

This reads like a nation state driven influence operation focused on feeding propaganda into LLM's and search engines (need to read towards the end to get to that part).

It's reasonable to expect stories the real local press finds discussion worthy (because they are both false and relevant to the local press) are an effective way of using the local press to throw more link strength at their own site.

>Today, 65% of our product team’s code is created by our internal version of Claude Tag.

Yeah, that explains a lot.

>GitHub is extremely insistent its employees maximally use AI for internal development

Or it could be that GitHub saw a 14x increase in commit volume last year[0], and we've concomitantly seen its reliability fall of a cliff in the last year or so. Given that Microsoft is leasing additional space on AWS(!)[1] to handle the additional commit volume, my personal money is on commit volume growth being a bigger issue than internal use of AI.

Internal use of AI may have been an issue. Commit volume growth may have been an issue. Unless one has direct knowledge of their infrastructure issues, claiming to know is quite literally making exactly the "they are vegan, their illness must be caused by their veganism" argument the GP commenter was talking about.

[0]https://daringfireball.net/linked/2026/05/04/commits-on-gith...

[1]https://www.businessinsider.com/microsoft-github-amazon-ai-c...

>Walt Disney Company is the most successful at monetizing human nostalgia.

Coca-Cola has a much larger market cap than Disney, and the Coca-Cola brand is very intentionally a nostalgia-driven, golden age, remember the good times brand.

It's great that you understand the chart. Yellow being more optimal than green is very far from intuitive to the rest of us. We read green then yellow then red as a sign of DECREASING fit. And a bunch of blues next to that green to red progression reads as "no data" or "not relevant" or something.

It's great that you have the data. I'm sure that took a lot of time to obtain. Spending a few more percent of your total time making the presentation of the data intuitive to others is almost certainly the highest ROI thing you can do at this point, if you want your site to be useful enough to get enough visitors to pay for the cost of acquiring the data.

Feel free to tell me that red is the lowest Pareto value and that's why you made the frontier red, or whatever. I'll still respond that the details of the presentation matters enormously to adoption, and the current presentation is very far from optimally intuitive to those of us who didn't personally develop the data.

It would help if you actually explained what the color means.

What is yellow? What is green? What is blue? Are they relative to their CPU column? Relative to the pricing row? Absolute?

>Not having a simple container based compute piece made me hesitate in taking up CF

Agreed. I wish CF had something like Azure's new fast-starting Express containers.

Calculations like this article neglect the far more common case where the liquidity event, if it ever happens, is years in the future, much longer than average employment times.

Companies used to go from garage to IPO rapidly. Now a decade from founding to IPO is not uncommon. Unless the employee plans on committing their life to the startup that just hired them, they will most likely move on somewhere else before that decade-later liquidity event happens.

When they leave, they have to spend after-tax $ to buy non-liquid shares of a still-high-volitility pre-IPO company, because almost all options have a 90 day exercise clause. If they were early enough that their shares are 409a valued at pre-seed valuation, that might make sense. If they are a significant employee, the kind of employee the company most wants to retain, and their shares are valued at Series B or C, that tends to be a very big check they'd need to write to their former employer to turn those options into non-liquid shares that may well end up at zero valuation, all at a time when they no longer have any ability to help the company succeed.

The original motivation for offering options was to encourage employees to stay, but what happens is the precise moment the company most wants those options to do their thing and retain the key employee is the moment when the employee starts doing the above calculation, realizing the company is years from IPO, it's still risky even if it looks promising, and the options are actually worth zero unless the employee either (a) stays at the company for a decade or (b) pays back a non-trivial fraction of their takehome pay to buy a risky illiquid asset. The employee realizes neither of those scenarios are likely to happen, so boom the moment when the company wants the options to drive retention ends up being the moment when the employee realizes the options are not just worthless, they could actually have a negative EV if exercised. Suddenly those options aren't very effective at retaining the employee.

There was a time when options had value like the OP suggests, because the timescales were shorter. Today, not so much. Yes, outliers happen, but in general they don't.

If you're not aware of "sync rights", it's probably worth reading up on given your interests. There is an entire specialization of music copyright law focused solely on synchronization of music to visuals. The good news is that studios almost never obtain this set of rights to the music they publish (because historically there wasn't enough money in it to justify negotiating for them).

I'd be more interested in a $29/month surcharge if Waymo weren't already significantly more expensive than Uber/Lyft to begin with.

Feels important, but I wish they also had compared against something like MeiliSearch or Algolia.

One of the links in Schneier's article makes a credible sounding case that Anthropic is using open source vulnerabilities as a shakedown operation.

When the model finds a vulnerability, it also finds a fix. Anthropic only shares the vulnerability with the Open Source maintainer, not the fix. Paying customers get fixes, confirming that the model does generate fixes for the vulnerabilities.

Sharing the vulnerabilities but not the fixes does sound like a shakedown operation.

Seems like there are some trademark issues with just calling this SQLite.

Naive question: what would I search for to find a tutorial on how to detect this on my devices, which are mostly iOS, or in my home network?

I'd love to find and remove any apps from my devices that have this SDk active.

Don't Microsoft and others already offer this?

> Adafruit’s reporting concerns a matter of public security interest and was conducted in the ordinary course of responsible disclosure

Very cool - I'd love to hear Andy ("greatest sales deck I've ever seen" blog post author) Raskin chime in as part of that conversation.

Real question, not intentionally meant from a tinfoil hat perspective: now that it's been shown the censorship can be viewed, how long before we see serious obfuscation of censorship circuits in LLMs?

>Everything was parallelized on Burla, on a single dynamic cluster that scaled to ~1.7K CPU workers for photo download and CLIP, with 20 A100 GPUs running embedding clusters in parallel on the same cluster.

That's a lot of budget - would have been nice if they'd made an actual donation to the project, instead of pounding the project's servers and bandwidth when there are much better ways to interact with the data.

Server load issues? Home page loads. Individual states don't seem to.

GitHub is a huge and almost 20 year old company suddenly experiencing massive scale growth as a result of an externality it didn't cause and that no one predicted. That is an incredibly difficult scenario for any long-running, established organization to handle.

Yes, GitHub is temporarily breaking under the increased load, yes, it's likely to still be a thing in 2 months, and no, it's unlikely to still be a thing in 12 months.

It's very unlikely a cool new thing will peel enough developers off GitHub in the next six months to survive long term as GitHub inevitably gets its ability to handle the new normal scale back.

>it should be pretty clear that if you provided the tool the specification for the code you want, you have already provided creative input.

If you provided a human contractor with the specifications for the code you want, the courts have repeatedly made clear you have not provided the creative input from a copyright perspective, and the contractor needs to explicitly assign those rights to you if want to own the copyright on the code.

The most important entry I found in my physical copy of the 1911 Britannica is for Eavesdropping[0], detailing the original historical origins of the term and how it was thought about just before our modern era.

> Though the offence of eavesdropping still exists at common law, there is no modern instance of a prosecution or indictment.

Thanks for posting this resource, I've often wanted to share a link to this and other entries.

[0]https://britannica11.org/article/08-0867-eavesdrip/eavesdrip...

So excited to see this - the big advantage of 1.58 bits is there are no multiplications at inference time, so you can run them on radically simpler and cheaper hardware.