softwaredoug

So this is like OpenClaw but managed and first class?

The big takeaway is they announced decision model API for Luna

Well I often forget the rules, so its useful :)

I’m reading The Well of Ascension, the second book of Brandon Sanderson’s Mistborn trilogy

I like how he has the slow, almost clinical way of revealing interesting worlds and systems. It’s not the kind of book that’ll make me cry, but I really like world building with fun enough characters

Yeah we forgot the goal of programming isn’t just to tell the computer what to do, it’s to program the programmer into thinking a deeper understanding of the problem.

Strangely I’d expect urban centers with apartments and street parking to be the problem.

In the US Electric vehicles are easier for those who can park a car next to an outlet they own. We have a lot more suburbia with people owning their own charging.

A lot of the hate is in the same spirit of the infamous hacker news comment discounting Dropbox because you can just setup an FTP server to do the same thing[1]

In this case, its people who have built BERT classifiers for years, pointing out you can just fine-tune some model to classify. Of course that's true, and still have value, but its not exactly turnkey. And Jev is surprisingly high quality at what its built for.

1 - https://news.ycombinator.com/item?id=9224

I don’t know if this is 100% true because there’s value in writing code for education and idea exploration.

I find it saves tokens and reduces slop piled on slop to have a foundational technical grasp of the technology behind your projects. And that means writing code for exploration and scaffolding.

Agents are such rich BS machines that I also don’t trust them. So I frequently debug and dive into my code rather than letting an agent layer brittle fix on top of brittle fix

My lack of stress about AI coding comes from years of being a principal engineer. Where the job isn’t to write 100% of the code, it’s to set technical direction and ensure we’re going in the right direction as a project.

I both write code and use coding agents. Because I find it’s often easier to set technical direction by showing some scaffolding, then later letting the agent take over the details. I waste less time with my agent if I can show it roughly what I want.

Occasionally I get disconnected by what my agent colleague is doing (or it gaslights me) and I have to dive in to right the ship a bit.

It’s not just price and speed, the API is very well designed for classification. Going beyond the current structured outputs.

Absolutely. I write a lot of code in 2026. You've got to be able to reason about multiple abstractions, and writing is thinking. In many situations, that means getting down to the code.

We have copious evidence for behaviors like note taking and writing creates deeper thinking and retention that passively watching / reading

https://www.scientificamerican.com/article/why-writing-by-ha...

> Unfortunately, the manufacturing environment has changed dramatically. The effects of COVID, increasing manufacturing and operating costs, and significant changes in consumer purchasing habits have made it increasingly difficult for a small American manufacturer such as Sherline to maintain the workforce and production levels necessary to remain competitive.

Can’t help but wonder if “operating costs” include tariffs and other policy decisions by the current administration that have driven up costs for everyone.

Is it really “research” - as in expanding human knowledge - if nobody understands it? The point is deepening human understanding, not producing research papers

We’re going to learn a painful lesson

AI isn’t responsible. People are responsible.

It doesn’t matter if it’s code, writing, or military decisions. People can / should be held accountable. As soon as people choose to remove their own accountability, that’s when the bad stuff happens. Whether it’s slop code or innocent civilians killed in a missile strike

This case is about a “visual search” as in CBP agent going through your photos. Not forensic tools copying / analyzing all the data

Doing a bit of research, this applies to what’s considered a “manual search” as in a CBP agent spending time manually looking at your phone. It does not appear to allow (or speak to?) plugging your phone into forensic tools for a deeper inspection.

Not great, but I think that’s what’s being specifically adjudicated here.

They still cannot deny an US citizen entry if you decline to unlock your phone, though they may confiscate your device. I don’t know if they would then be allowed to do deeper forensics without a warrant.

It’s not about what the AI can do, it’s about your ability as a human to keep up with the details so you can make effective decisions. You the human are ultimately responsible.

Because where dangers lie is when you lose track of that one tiny important thing that turns out to really really matter.

The AI won’t get fired, you will.

Thank you!

In my work on LLM as a judge, I prefer to use LLM decisions as features in a downstream classic ML model for the final decision. It works really well

https://softwaredoug.com/blog/2025/01/21/llm-judge-decision-...

“Using and keeping secret” - for nefarious purposes? Or to make Linux work on PS5?

> The scene used to be a group of highly talented researchers, but now it is just a bunch of noobs using LLMs and writing hacks they don't even understand. Slop kiddies found the only hypervisor bug left, which I had also found a while ago, and decided to report to Sony.

I don’t know if he’s mad AI scooped a bug? Or that LLM slop kiddies don’t pay attention to that something is already known?

That’s obviously reductive. They don’t have to be forced. They could mutually agree to unify more strongly

I’m just telling you the tradeoffs you’ve chosen. You can’t be an economic superpower in the current regime.

Without your own “hypergrowth” you’ll always be dominated by the hypergrowth of competitors like the US and China. You’ll be in orbit of other powers instead of creating your own gravity.

Where am I wrong?

Ok but you have to accept that means an EU country works with a hand tied behind its back.

In the US I can start a company and instantly have a market of 350M. Start a company in an EU country, then try to expand to another country, it’s a nightmare. That company will have a lower upper bound on market size.

The EU might be a success in coordinating on regional issues, but I don’t think it’s yet an asset to the European economy.

Absolutely not.

I have way more hard problems that engage me then ever.

I do both human and AI coding. The former more to learn and mold my mind around a problem. The latter when I feel I am ready to delegate to a junior dev.

I think of AI coding like how I’ve operated as a principal engineer at a company. It works well when you study the problem well enough to know precisely what you want. That often requires actual hands on technical work IMO. The coding isn’t for the machine, its not even to ship, it’s to help me think. Then I have a basis to really understand and care about the details. I can micromanage agents to do what’s expected at this point.

I want this one to succeed, but refactors often fail and revert to the original code structure :)

Maybe, but it’s way too difficult to do business in the EU.

Not because I think regulations are bad. But regulations need to be well designed with clears lines on what the Federal / State are allowed to do.

I’m not sure this is a good thing for Canada.

The EU feels like an incomplete experiment. Not strong enough to actually be a single market. But still able to create additional annoying regulations on top of nation level ones.

The EU needs to be truly Federal to succeed economically. That might create uniform regulations with more uniform regulations and actually make Europe competitive.

FWIW, Naza appears to be a fairly in depth documentary where Israeli soldiers give detailed accounts on how the Gaza war worked from the inside, including expansive targets, and well known high civilian death toll.