lostdog

I should set up a LARP where 30 people solve TIS-100 together

But lately there's capital available for quantum computers, fusion, synthetic bio, space exploration, asteroid mining, and lots more

Hmm I wonder how these were originally funded when it was less likely that they would work.

And by cutting funding now, I wonder what we're missing out on in the future.

I have some coworkers that are similar in everything--education, work ethic, and intelligence--but some of the tick out ML ideas that work like clockwork, while others get hits rarely if ever. I cannot tell what makes it work for some and not others. Their ideas both sound equally good.

Sometimes a coworker will be an ML star for a year or two, but then suddenly run out of steam. It's brutal to watch.

I used to think most smart people had similar distributions of good ideas, and it was just that the hardest working tried out all 50 of their ideas to pick out the 2 good ones. But I've seen smart and hardworking people have a hit rate of 0.

1850s was the beginning of the railroad boom. You're thinking 1750s.

"a real balance sheet, assets against liabilities, not a spending feed dressed up with colours"

Ugh. Just write your own damn post already

C++ was a superb language for its time. There was nothing faster with as-powerful abstractions. It showed how far you can change a language too, with C++11 being a massively better language with shared_ptr and company.

It took in almost every idea, and the battlefield showed us which do work and which don't. We get to keep RAII, move vs copy, smart pointers, placement-new, and generics. We get to drop auto_ptr, copy-by-default, its specific exceptions implementation (fight me), multiple virtual inheritance, and templates as full code substitution.

In my opinion the battles have played out, and Rust is the best sum-up of what worked (it even inherited the compile times! Lucky us!)

AI code is often fine. The point of the code is for the computer to do a job. And code is supposed to be consistent and boring and straightforward in its structure. AI is pretty ok at this.

AI is terrible at writing. Its prose is awful and horrendous to read. Plus, the whole point of reading a written piece is to hear about the author's new ideas or experiences. If AI wrote your piece it's both bad and pointless. I can also prompt the AI to hear what it knows about.

I haven't tried opus 4.8 yet, but I hope the writing quality has returned to the Opus 4.5 level. Anthropic really lost something, where 4.5 had this really crisp writing style that flowed really nicely and 4.6 and 4.7 sound much more "chatgpt-like." It feels like they tuned it to be too much of a problem solver, and when you do that you get this terse, clipped textual output that's more difficult to read.

It is because they have power, and want to be addressed this way. And you are likely to lose your case if you don't follow their rules.

The major axis is urban+gritty to more suburby and spread out. It's a very personal preference where you want to be, but most people dislike the most gritty areas (tenderloin, most of soma). It's worth aiming for a neighborhood at the median as your first.

Yes, bbc had gone back to paywalling this month.

Is real, physical engineering that much better?

Take the new sf bay bridge span. It leaked, and had to be fixed to prevent critical parts from corroding. https://www.sfgate.com/bayarea/article/Caltrans-was-warned-o...

Projects are consistently over budget, late, and shoddily done in the physical world too.

Yes, if you just list 3 more problems about the US then it means that China has no problems at all.

Grok is despised because it has more aggressive alignment.

The best managers I've seen would turn this situation into a headcount request.

The problem is leadership has priorities 1-5. Your team works on 1-3, but the PM keeps getting hassled about 4 and 5, so they look for levers to get them to happen.

In this situation, the PM scrounged up headcount from elsewhere, but if you present the option of adding headcount to the existing team, then you create a more harmonious option of getting these lower priorities accomplished.

Of course, this guy was taken fully by surprise by the suggestion. It's much harder to present a better option after the fact, and I agree that letting leadership feel the consequences of its decisions is a reasonable thing to do in this case.

And uv required some ground work, where the PEP process streamlined how you define a python project, and then uv could be built on top.

Is there something mass-produced that's flexible and consistent like gridfinity? As a non 3d printer owner, I've been looking for something I can just buy that would work.

Maybe it's time to do pair agentic engineering? Have two engineers at the screen, writing the prompts together, and deciding how to verify the results.

Whoosh

In MCP setups you do give the agent the full description of what the tool can do, but I don't see why you couldn't do the same for executables. Something like injecting `tool_exe --agent-usage` into the prompt at startup.

Great article otherwise. I've been wondering why people are so zealous about MCP vs executable tools, and it looks like it's just tradeoffs between implementation differences to me.

Yes, sites should support a NoAds header that agents can provide, which ensures that the site doesn't provide any ads that the agent could accidentally click on.

And get it into a modern certification. Want LEEDS? Get the sound measurement people out.

Please post your address. I'd like to help make your home "feel alive."

They are oppressed by their neighbors, who can scribble all over their home without consequences.

Have you had to clean off graffiti?

We begin teaching math by having students solve problems that are trivial for a calculator.

Though I also wonder what advanced CS classes should look like. If they agent can code nearly anything, what project would challenge student+agent and teach the student how to accomplish CS fundamentals with modern tools.

I wonder if you could calculate a "probability of response to major alert" and make it the inverse of the total or irrelevant alerts. Then you get to ask "our probability of major alert saliency is onlt 6%. Why have the providers set it at this level, and what can we do to raise it?"

Is it ironic that the project with the most potential for making C++ build quickly is suffering the most from C++ being extremely slow to build?

Beyond that, what a superb overview. More open source projects could learn from this posting. It provides a great list of areas for someone who wanted to get started contributing.

Weird. I find Opus knows the answer more often, plus its explanations are much clearer. Opus puts the main point at the top, while Gemini wanders around for a while before telling you what you need.

There's a simpler design here begging to show itself.

We're trying to orchestrate a horde of agents. The workers (polecats?) are the main problem solvers. Now you need a top level agent (mayor) to breakdown the problem and delegate work, and then a merger to resolve conflicts in the resulting code (refinery). Sometimes agents get stuck and need encouragement.

The molecules stuff confused me, but I think they're just "policy docs," checklists to do common tasks.

But this is baby stuff. Only one level of hierarchy? Show me a design for your VP agent and I'll be impressed for real.

The right number of frameworks in a company is log10(number of employees). Go over this limit, and everyone needs to learn too many frameworks with too little support, examples, tools, and documentation.