benchwright

You said: "and support expectations clear,"

Therein lies the rub. By not honouring the trademark, the fork made the association of service, support, otherwise to Notepad++ making it seem like it was officially supported.

Imagine if someone who used the fork attempt to get support on a product that wasn't supported and, when faced with limited responsiveness, etc. decided to denigrate the original developer by lambasting them on HackerNews, et al. The reputational damage alone would be seen as a reason to defend the mark.

I work with a very large OSS nonprofit who has trademarks in most of the geos around the world and vigourously defends them for precisely this reason: reputational damage undercuts the community, the developers, and the reason for existence.

In the grand scheme of things, this isn't shocking. The push for OSS weights and outputs (looking at you, GPT-OSS) is great and all but the SOTA still sits within proprietary, closed-source land.

Can agree with this (to a certain extent). Spending time in both America and Ireland, there's a definite difference in the extent to which "I can only get there by car versus public transit/other types of transport" is incorporated into the culture. By far, more urban/suburban/rural density with appropriate public transportation support leads to more freedom to walk, bike, etc. Geographically, however, there are clear reasons why places like the US and Brazil have a strong incentive to rely on cars over anything else. Distance between cities and towns, poor public transport (esp. outside of urban areas in the US), etc. cause this to be an issue.

That's not a transaction fee, that's dropped change.

almost like NOAA, had it not been savaged by this current anti-science admin, would be able to lend further credibility here. As it stands, several researchers that I know of, formerly of NOAA, were banging this drum loudly for decades.

So, curious application especially given the work that Benchwright.io, MLCommons, ArizeAI, and others have provide regarding drift, etc. How do you differentiate yourself from those offerings (knowing I'm not neutral here but fascinated by drift, tolerances, characterization, etc.)?

One of my customers actually requires attribution to agents if they're used, not only for tracking purposes but also for understanding potential vectors for slopcode. It's been useful and occasionally enforced. That being said, implementation without due consideration and warning should be frowned upon.

Great article and something that I've envisioned myself: a forge combined with a libary (of sorts). Looking forward to the forcing-function of GitHub's changes recently pushing folks to develop new, more engaged ways of maintaining code.

I managed to build this the other day. Looking good!

As noted by other authors, there does seem to be a wide surface area for attacks so, security remediation does seem to be a top-of-queue thing to be handled.

All in all, keep up the good work!

Tends to be a problem. I've tried to mitigate these problems by using either external harnesses (aka GitHub actions that are "fixed" based on known-good) or by using n-number of witness agents (e.g. Kimi/Qwen/whatever <=> Claude/OpenAI/Google). Generally sucks more time and energy (and now token/$).

that being said, I still have a "fix the code, not the test" line somewhere in here...

I don't see that at all. There is certainly a multi-causality/multi-maintenance problem here but in a grossly capitalist market, Spirit's ability to thrive was curtailed not only by social presence (they're the Jet2 of America) but by an inability to function in a lean business model like Ryan Air (an exemplar of what Spirit Airlines COULD have been). M&A isn't always the appropriate exit mechanism. We have Breeze (small market), Frontier, and while Southwest has certainly exited the "value" end of the market, it still dips its toes in the water. Also, the prime carriers "Basic" tickets certainly ape what Spirit tried to do...albeit with better access and scheduling.

"Bad" is relative. It can be more difficult in compressed markets where talent is either in surplus (read: SV, Dub, etc) or where there's a distinct lack of enterprises/startups/whatever providing the surface layer. Given some of the retraction of "warm feelings", esp. re: US contractors in other countries circa American imperialism these days, it can also have a chilling effect on localized markets. But, this is all highly influenced/mutable daily.

PII in plain text is a problem. Glad there's a level of accountability here.

They can be. I've done some drift detection work against local models and for the most part, they do ok. I think there's always room for an augmented approach where local models can handle programmatic parsing and structure and large models can handle actual coding routines. I try to use witness coding using local+api where possible to see if there's capabilities that can be caught one side or the other.

https://www.transportation.gov/briefing-room/trumps-transpor...