SOLAR_FIELDS

Mind unwrapping this a bit more? I’m interested in iops heavy distributed caching and this seems like it unlocks some creative use cases there

I look forward to the evolution of your project into a less standardized Kubernetes as end users request more and more features of your project.

The biggest problem, IMO, is that if you buy from a third party when shit hits the fan you have to rely on that third party. This is still a problem even with reputable travel agents like Amex and Chase. Even in the best case scenario where the agent is actually good, having to get a flight changed via a phone call with some agent who might or might not answer promptly and might or might not be able to work quick enough to win a race of rebooking after a cancelled flight before next best one fills up is just asking for a horrible experience. If you book with the airline directly they have a lot more leeway and power to unfuck your situation promptly.

Save your points and use them on hotels instead, where the experience is just as risky, but at least fails in a less spectacular way when it goes wrong e.g. unless you're booking with an agent in a high demand area where there are NO hotels you usually can leverage a backup plan a lot easier than if you are stuck in an airport.

I was big into dumbphones about 10-15 years ago and the problem with dumbphones is the same problem with dumbtv and other ones, is that the market is already small, and those in the market for these things are opinionated as shit about the specific configuration they want. So you are presented probably with maybe 2, 3 viable options if you are lucky, none are the thing that satisfies your actual needs, and they are all overpriced as shit because they have no competition because the market is so niche. So you probably end up buying the closest configuration possible to what you want and then spend your experience being slightly annoyed that it’s not fully there.

Austin, Texas is a great example for toll lanes of existing roads! MoPac and 183 North are exactly this model

Agree that it's a more accurate metaphor than just "toll roads".

My understanding is that Cintra exited the Austin TX toll road business around a decade ago after profits from SH 130 failed to materialize

So it was unintentional behavior in a beta release? Yeah that definitely does feel like something we should be getting up in arms about

Australia is one of the top toll roading countries. Sydney had most tolled roads world record for awhile not sure who holds it now. I don’t know whether it is the state or private companies who operate them.

Since you live in the area, and are talking about toll roads, I assume you also probably know of the name Cintra, who are Spainiards. And Cintra is, in fact, a private toll road operator who has owned roads around Austin previously

I do find it amusing when I see these LinkedIn fodder posts saying “I left [Meta/Google/OtherGiantCorp] because the company I joined 10 years ago doesn’t hold the same values anymore” as if these companies weren’t already fucked up evil pieces of shit many years before you joined. People are always trying to justify that they aren’t working for something purely evil. I guess you have to try to feel like that giant pile of cash on offer isn’t dirty money and you need validation from others to sleep at night

This is pretty close to how toll roads work tbh

I don't understand why this is being presented as an either/or thing.

The moat is actually the harness AND the model, and one of the reasons that Claude works so well is because the model is actually trained with its usage in that specific harness in mind, and the harness is designed to deal with Claude model's idiosyncracies. Easy to validate, just run Claude through some other harness and compare, then just run some other model through Claude's harness and compare

I’ve been running an SMB trending mid market out of gitlab for the greater part of a year with no issues. One of my favorite benefits is that ci runners are colocated with the self hosted instance on k8s so suddenly a whole huge slew of shit that you had to care about with GitHub, security, provenance and supply chain is just… not an issue.

Getting off the GitHub actions dependency is a feature, not a bug

There is a 5 line skill I’ve been using for refinement called grill-me that works quite well

Of course, people actually good at security are rare and in high demand. This is totally aligned with OP’s statement. IMO you shouldn’t even be thinking of going into cybersecurity straight out of college. There’s just too much you have to learn about how software works for it to be a reasonable first job out of university. There will always be exceptional people, of course, but as a general rule I’m not hiring new grad cyber folks. Seems dumb

I personally as a general rule don’t hire people who work in cybersecurity if they were not traditional developers first. The chances of you understanding “cybersecurity” without also understanding how general software works is extremely low.

> It is fundamentally important that convential commit is better for adopting it to be an improvement.

Absolutely! Which means we can also agree that conventional commit is objectively better than No System For Their Commits At All, which is what 99.9% of people are actually choosing between when evaluating conventional commit. They aren’t looking at conventional commit vs Some Better Way, they are looking at “we have no standardization of commit messages” vs “we have standardization of commit messages”.

For the 0.1% people for whom that’s not good enough, one hundred percent agree that these people should be pursuing better solutions.

Indeed, but circa 2019 the max taxable amount that the value of the home can rise any given year is capped in Texas. So if you are in that situation yes eventually the taxes will “catch up” and you will pay more every year, but that catch up is capped every year and you know what it is capped to and can plan for it

“A ton” is carrying a lot of unverified weight here. Do you have demonstrable evidence that move rates are higher in Texas AND that it’s a result of this AND that people hit this with an actual interesting frequency?

True, but in my experience grandparent is a bit full of it. Yes property taxes are higher, but what op leaves out is that the values of homes are like for like 2-3x more expensive in California. So in the end I would probably pay around the same tax for a like for like home in California as Texas, simply because of the value of the home. Then we must consider the state income tax of California, which is a nonzero differential to that of Texas’ state income tax

It’s the linting argument all over again. I don’t give a shit what the style is. I give a shit that it’s consistent. Form whatever opinion you want about how you want to format your code, structure commit messages etc. I don’t really care, if you want to start every commit with “poop(fix): pooper my commit message”, as long as you’re consistent about it and enforce it programmatically you have my emphatic support

This comment assumes you are picking between a conventional commit message or something better. But the reality is you are almost always picking between a conventional commit and Nothing.

My anecdata from my example demonstrates it’s not the case. I hit the security guardrail, then start a new prompt, asking it to do literally the exact same thing in a different way and without the lead up context, and it happily does it

Not directly, as it comes in as a not charged error but the weighted generation path used until you hit the guardrail is basically wasted tokens, so yes, indirectly. If I hit a guardrail and rewind I’ve found the training will still be biased towards guardrailing out if you rewind one turn. Rewinding multiple turns allows steering away from that path, but all of the original token spend down that path is wasted

One interesting takeaway is the low score on Anthropic models from this benchmark. It’s not because of capability, it’s because Anthropic’s guardrails prevented it from solving the problem.

I noticed with each model release Anthropic constrains the model more security wise. Its propensity to refuse doing legitimate work has been increasing. It now puts up more resistance around performing logins, handling credentials on behalf of the user, etc.

For myself, it’s already gotten to the point where it has mildly affected the usefulness of the model. If I bump on some action I want it to do I can usually work around it, but I suspice the ability to do so will close with each new release. Eventually I’ll reach a point where I am forced to choose between the useful aspects of the model and the limiting ones instead of just picking the most capable model out there

Eventually these models will significantly suffer from overfitting to the least common denominator. If I have this beautiful deterministic setup that swaps secrets out in flight so the LLM never sees them, I’m going to be really annoyed when the LLM still won’t send them out because it is trained to deal with the 99% of people just doing the dumb thing

If your product becomes commoditized, it’s no longer a high margin business

Actually yes, when it’s other huge corporations holding them accountable. It’s only when politicians who are much more cheaply bought get involved that creates problems. When the other side has a significant war chest to combat you with, suddenly behavior improves

They aren’t and no one legitimately did. This is extrajudicial surveillance state stuff

Limited sample size but I've not worked anywhere in the US that didn't have a 90 day carve out for terminating people from date of hire, MegaCorp or SMB.

Of course it's not cost effective, or efficient, but as the article points out, nothing really is, and this method is at least preferable to the weird quasi-employment situation the article proposes

I was being facetious, sorry if that wasn’t clear. That’s just how Regular Employment works. You don’t even need to tell the employee the implications of being on the 90 day plan. You tell them up front what the expectations are for the role, and if they don’t meet them in 90 days, you fire them. Does not need to be more complicated than that.

30/60/90 is pretty standard stuff. It’s literally the documentation and justification you need to fire the employee at the 90 day mark if you need to. Like a preemptive PIP if you will.

I don’t want to hire anyone who doesn’t want to be held accountable to what they say they can do anyway.

Instead of having “provisional employment”, could we instead just have 30/60/90 day plans and evaluate the performance of the engineer and fire them if they don’t meet that criteria at the 90 day mark?

If I get a group of 30 kids together that are incredibly intelligent and highly motivated and have had “you must be the best and you must get A” beaten into their success and livelihood since before they could talk by their parents (and let’s be real that is a good chunk of Harvard grads) - do you really think that telling them that they are going to be stack ranked against each other is a healthy and productive thing that will produce the best outcome?

The thing about the framework 12 is that they are giving you an open device that is meant to be upgraded. The value of that is different to everyone, but to place it side by side hardware wise and try to compare it as if it is equivalent on the software side to the closed source bullshit Apple has on offer feels at the very least a false equivalency