HN user

MarkMarine

2,943 karma
Posts0
Comments793
View on HN
No posts found.

If you haven’t talked to Anthropic support yet you’re in for a surprise. I’m an engineer at a company with an enterprise contract, Anthropic people in our slack and it took me a month to get a response on my support request, I just decided it wasn’t worth it and bought a second phone number rather than wait.

Performance. A safe Swiss Army knife function that did most of this would be slow because of the internal branching you’d need to be safe, and because there is developer intent in the selection of these functions. I’d rather have the choice and clear dev intent when I see the function used when reading code.

I don’t see it that way. Some see the invisible hand of the market but I just don’t, the involvement in politics… the purposeful shaping of public opinion that is against the public interest, capital has an interest and it’s making its interest felt.

I’m pretty sure the burning fossil fuel and mowing down rainforest isn’t in the long term interest, I’m also sure that if I’m dumb enough to see this the people in control are as well. So I admit I can’t see the end state, but I’m sure the machine is being fed somehow

The whole framing you’re using is wrong. If you can do climate modeling and math at the planet scale, you don’t need to be convinced you have read the research and you knew 30 years ago.

If you’re skeptical of scientific authority and lack those skills… as well as the critical thinking required to read the research and distill your own conclusions, there is basically no way to convince you. You can’t find objective truth yourself and you don’t trust the resources that can.

Doesn’t matter anyway, you as an ant in a large colony barely matter, no one needs to convince you. Capital needs to understand it’s in its interest to fix the problem and it will be fixed. Until that, get an A/C and contribute to the problem.

My go to for these (was a tractor mechanic early in life) is to start with a big flat blade screwdriver and knock it completely through the side of the filter to the other side with a hammer, then use that to break the filter loose. I’ve near had one go sideways with that method, you get it through both outside sheet metal and the inner perforated metal that is usually stronger and welded to the threads

Definitely got attention during the prod outage at my work. I’m going to find another alternative, I’m sick of this terrible uptime.

This type of pedantry doesn’t add to the discussion. You’re correcting someone’s spelling on the internet, it’s pointless except to make you feel superior, adds nothing. Make a salient point, bring something to the discussion please. Sorry I hurt your feelings bud.

One of the main problems I have with the models coding is the feedback loop is way down the chain from generation, it's out at the commit boundary for python when your hooks are running, maybe at the point where the model wants to push a PR. The REPL lets that happen during generation, and the other safety measures help immensely. Immutable data, STM, all of the features in Clojure that gave devs super powers now do the same for a model.

Claude Opus 4.7 3 months ago

Anthropic was chirping about Chinese model companies distilling Claude with the thinking traces, and then the thinking traces started to disappear. Looks like the output product and our understanding has been negatively affected but that pales in comparison with protecting the IP of the model I guess.

As an aviator, that right there counts as an emergency landing. A hard one.

They limped it back home, they didn’t ditch a very sensitive airframe over enemy territory, I’d call that a win and the pilot deserves a medal for that.

I don't have a good answer for you. I expected the upper and middle officer corps to conduct themselves with honor and they aren't.

I'm going to bet that pilots aren't briefed to hit a school, they get a target package that says this is a legit target, an IRGC command post or something. There are multiple layers of detachment between the person picking coordinates, entering them into a JDAM, and the pilot releasing that weapon so who is ultimately responsible (and this is by design, everyone can tell themselves a story right now to sleep at night.)

But you do know what you hit, in the version of the military that I was in there would have been a detailed investigation into the chain of failures that led to striking a school with children in it. I'm sure it weighs heavily on the every person involved in that decision. Cold comfort for the parents of those kids, but something like that leaves a life long scar on the people responsible.

I know you're not.

I've found that most of our population has almost no connection to the people that actually fight wars, and therefore have no idea what they think. With the exception of a few criminals, none of us desire to commit war crimes. None of us want to send rounds into civilian infrastructure, seeing regular people struggle to get food, fuel, and water in Iraq did not make me feel powerful and it was obvious it did not advance our goals on the ground.

The jingoistic commentary people hear from politicians and former military podcasters that don't fight anymore is repugnant, and this backsliding in the (at least attempt at) honorable execution of war is not going to bode well for our country. It's probably trite when we're double tapping girl's schools, but I want to think that purposely striking civilian infrastructure, universities, hospitals, water resources... this was all something "we" didn't do.

This is actively devaluing the meaning of being a Marine. Maybe this already happened in Mai Lai, maybe this was further chipped away by Abu Ghraib, maybe letting Eddie Gallagher off... etc etc. But this feels different in a way I've never felt before.

I didn’t downvote you, but a terse “well actually it’s prisoner of war” doesn’t really add to the conversation. Imagine doing that in person, you’d annoy everyone around you. If you explained why it’s distinct and what that might mean for downed crew I think it wouldn’t have been down voted

No, we actually train to be tortured and held if caught, but everyone knows the risks before you take off. Captured marines or soldiers have been killed in Iraq and Afghanistan, we’re clear eyed about it.

Military aviators train for this, being alone behind enemy lines (look up SERE school if you’re curious, one of the craziest training courses outside of special forces) and there is a special force just for aviator recovery behind enemy lines, US AirForce Pararescue. Hopefully they’ll get the aviators back quickly, the last thing our country needs is American hostages making this ridiculous war harder to stop.

It’s almost shocking that people in an era of unlimited resources could see this was not renewable and important to hold, and that later in the era of limited resources, we decided to privatize this. It’s so shortsighted, willfully ignorant.

We’re about to get a preview of the world after fossil fuel extraction and some of the knock on effects. Semi is one thing, wait till you can’t get an MRI.

Over a number of files similar to a codebase, that are well organized (like a codebase) the coding agents and harnesses are quite good at finding information, they clearly train on them so they will only improve.

The challenge is how to structure messy data as a filesystem the agent can use. That is a lot harder than querying a vector db for a semantic query.

The code bases we’ve been using agents in had been pruned and maintained over years, we’ve got principles like DRY that pushed us to put the answer in one place… implicitly building and maintaining that graph with all the actors in the system invested in maintaining this. This is not the case for messy data, so while I see the authors point and agree that a filesystem is a better structure for context over time, we haven’t supplanted search yet for non-code data.

OpenAI Symphony 5 months ago

I have a fork, it's excellent. Complementary to something like symphony which doesn't deliver the harness you need to make working software, attractor's graph orchestration inside the loop that symphony creates can set determinative workflows with the harness openAI (and StrongDM) were talking about using to test. My personal system uses property testing, fault injection and fuzzing layered on top of playwright and e2e testing behaviors against a digital twin system I have to credit StrongDM for inspiring again.

Yeah I do, first it’s a joke but second have you used Ironclaw? It’s buggy vibecoded software. Bring your Claude code with you to get it working if you try it. So yeah, the Primegean joke about adding “but make it secure” to your prompts is true about this, even if Illia is a genius and I look up to his work.

Ok, I glossed over color matching the wall patch. Fair.

But there really aren’t many walls you need to open in a house. There is probably 2-3 wet walls, so unless you need to retrofit some ducting why are you opening a wall? Code says there are no hidden wire junctions, so you’ve just got continuous runs of romex that are secured before they terminate… what do you open a wall for?

Most of the drywall repair is just physical damage to the drywall itself.

I think this misses the beauty of a plaster wall. Level 5 drywall has nothing on a skilled artisan with plaster, and yeah you can’t hang things through it but it also lasts hundreds of years. My walls are 120 years old and robust, the kids haven’t damaged them and they’ve more than held up.

Drywall is trivial to remove and repair, I have no issue cutting walls with a circular saw or vibrating cutter to get access then patching it.

I have seen another method for making walls that were accessible though, from a homesteader/ hand tool woodworker and carpenter. His walls were 24” thick with huge areas for piping and electrical and had 4x4’ removable wood panels.

https://youtu.be/8fdm9R1Cbm0?si=9SRXgcdutos-hywc