mulmboy

Yeah I get this. Often I'll prompt it like "my intern looked at this and said maybe you should x. What do you think?"

Seems to help.

Via https://simonwillison.net/2025/Mar/4/deepsearch-deepresearch...

I've seen similar things with claude and OCR with low temperature. Higher temperature, 0.8, resolved it for me. But I was using low temp for reproducibility so

I like it! It's provocative and opinionated, just what I want out of a blog. Blogs are casual, personal, subjective. I read a lot of blogs like this and appreciate being able to take in many broad opinions, many of which yeah I think are dumb, but I like to have them all bounce around my head. Gives me a window into how people are thinking across the community

what you're saying is orthogonal to the post I was replying to, and to my reply.

self interested and short sighted says nothing about intelligence, irrationality, or poor decision making. Back handed and inhumane covid hiring and firing is probably not a mistake from their perspective. professional bullshitting is a form of intelligence (I hate it too, I've been done in by it, but I respect it)

> I'm happy you've only worked for altruistic, not-for-profit minded companies that care about employee growth and takes pride in their tach stack above all else. I have not had as fortunate an experience.

No one is making this claim.

My comment was a bit terse and provocative, rude, deserves the downvotes tbh. I'll take them.

To elaborate ~ I've got a lot of empathy for the poster I was originally replying to. I've fallen into that way of thinking before, and it sure is comfortable. Of course, companies and their leadership make poor and irrational decisions. Often, however, it's easy to perceive their decisions as poor and irrational when you simply don't have the context they do. "Why would they x ?? if only y!!" but, you know, there may well be a good reason why that you aren't aware of, they may have different goals to you (which may well be selfish! and that doesn't make them irrational or anything). Feels similar to programmers hating when people say "can't you 'just' x" - well yes, but actually there's a mountain of additional considerations behind the scene that the person spouting "just" hasn't considered.

Is leadership unintelligent, or displaying poor/irrational decision making, if the company self destructs? Perhaps. But quite possibly not. They probably got a whole lot out of it. Different priorities.

Consider that leadership may label a developer unintelligent if that dev doesn't always consider how to drive shareholder value "gee they're so focused on increasing their salary not on business value". Well actually the dev is quite smart, from their own perspective. Same thing.

And if every company you've ever worked for truly has poor leadership then, yeah, it's probably worth reassessing how you interview. Do you need to dig deeper into the business? Do you just not have the market value to negotiate landing a job at a company with intelligent leadership?

So, two broad perspectives: either the poster has a challenge with perception, or they are poor at picking companies. Or perhaps the companies truly do have poor leadership but I think that unlikely. Hence it comes back to the individual.

@y-c-o-m-b sorry for being a bit rude.

Cheers for reading

> After 20 years in tech, I can't think of a single company I've worked for/with that would fit the profile of an "intelligent" company. All of them make poor and irrational decisions regularly. I think you over-estimate the intelligence of leadership whilst simultaneously under-estimating their greed and eventual ability to self-destruct.

Says nothing about companies and everything about you

> you also over-estimate the desire for developers to increase their productivity with AI. I use AI to reduce complexity and give me more breathing room, not to increase my output.

I'm the same. But I expect that once many begin to do this, there will be some who do use it for productivity and they will set the bar. Then people like you and I will either use it for productivity or fall behind.

Is processing one page at a time not feasible? I'm always chunking things as small as possible for LLMs

Not adding to the discussion but couldn't help myself

> Show me everywhere this API is used incorrectly across the entire codebase

This is possible these days with (you guessed it) LLMS, likely with RAG. In fact I've made MCPunk https://github.com/jurasofish/mcpunk with exactly this in mind. There's a nack to wielding it effectively (bullying the LLM just right, feeling when it's bullshitting, etc) but it feels like this kind of capability will soon be commoditised. Super handy for code review like "Hey pal are the changes in the currently checked out branch logically consistent with the rest of the codebase"

"properly" is the key word here. You've got to communicate what you want. Or at least communicate something like what your goal is from the diagram, why you want it, so the LLM knows the audience it's targeting.

Like imagine giving the same prompt (instruction, directive, task) to a human - you would in all likelihood get out a similar high level diagram because you've not provided even the slightest whiff of what you want to use the diagram for.

The blog's takeaway is essentially "LLM didn't read my mind so no good". They're tools to be used and you get out what you put in.

This is a poor quality blog. They "upload" source code to chatgpt so who knows if it's in context or ragd plus it looks like they reuse the chat from a previous prose -> code session. They criticise LLM for high level diagram despite using a garbage prompt and rejecting the idea of iterating on the diagram.

Anecdotally I've had great success with code to diagram via LLM including fine details. But as with anything LLM you need to really get the context right. This can not be overemphasized. And iterate with the LLM, goodness.

it's a stylistic choice, youthful & casual,,

I take a similar approach where I have branches named `my_feature/x` where I start with `my_feature/1`, do some stuff and make a messy commit history, then when I feel like I want to clean up the commit history I branch of that to `my_feature/2` where I rebase/squash/reorder/rewrite away and continue from there. That way if I lose something important I just go back to an earlier branch. for complex features I can easily get to 15 of these branches. And I just delete them when I'm done.

I find this works really nicely, I can periodically clean up my commit history and do a full review of my changes so far and make sure they're all coherent before continuing. And doing this regularly makes it much easier to keep the commits nicely self contained. It's like `final_version_final_ii_for_real_final.xlsx` on top of git

Amazing! I am half way through building exactly this (and was even toying with the exact same name) but this is better plus you've beaten me to release. I have a heap of ideas and will jump in on an issue or discussion later today, would love to ditch my work and adopt this

Yeah the article doesn't even mention "what about more threads". Responses to your comment suggest there's broadly a lack of clarity around what Async gets you especially vs threads.

If the alternative to Async is more processes, rather than threads, a clear benefit is reduced memory usage and reduced process startup time.

`if x: foo()` is a cancer on the Python community. Devs often use it with the intention of handling x being None, and carelessly lump in zero and empty lists/strings at the same time. Endless bugs.

These tools are neat and handy for simple diagrams. Love mermaid.

For more complex diagrams, where explainer notes are beneficial, I find hand tweaked diagrams can not be beaten (my go to is draw.io)

If a simple DaC diagram is 1000 words, then a hand tweaked diagram is 10,000. And that 10,000 pays off. The bang for buck payoff is there. Of course, other considerations like maintainability etc etc

Rust approach may be great, strikes me as potentially a natural fit for a spatial index though - curious if spatial index would be practical and how it compares. Geopandas from memory has at least some support for spatial indexes. I believe geopandas if used naively will also do spatial operations quite naively (slowly)

> P.S. Here's a nice little tidbit in the source:

Nice

Bello, fellow Minion! Let me tell you about the banana-tastic article I just read called "LIKE SLUGS THROUGH THE HOURGLASS, THESE ARE THE DAYS OF OUR TECH TRIUMPHS"! Minion 1: Ooh, banana! Tell me more! Minion 2: Bananaaaa! What's it about? Minion 1: Well, it's like a big banana split of tech job market woes! The writer talks about how interest rates going up made tech jobs go splat, like mushy bananas. Minion 2: Papoy! That sounds bad for techie bananas! Minion 1: Totally! And get this - some companies want one Minion to do the work of many Minions! They call it "Everything Bagel" jobs, but it's more like asking for a banana, apple, and potato all in one fruit! Minion 2: Bapple? Bapple bad!

> Additionally, Petr set up the first project homepage for Git, git.or.cz, and a code hosting service, repo.or.cz. These websites were the “official” Git sites until GitHub took over.

Is this true? I thought GitHub had no official affiliation with the git project

> Failing tests indicate the presence of bugs, but passing tests do not promise their absence.

If only :)

Far too often I find myself working with tests that patch one too many implementation details, putting me in a refactoring pickle

sure it's casual but the things you're pointing out are merely an alternative style and don't suggest lack of maturity. judging a book by its cover

Is self hosting possible? Can't see any mention in the docs

Seems it wants "Act on your behalf" permission for GitHub which is a non-starter for me. Pretty neat on public repos though.

I find it's interpretating the intention of and figuring out the unhandled edge cases of an existing regex that drives me away from them. A tool you can ask "Hey how are you? Given this regex and this code context how will this fail?" would be valuable (always have to be nice to the AI)

How are JSON operations not atomic? Genuinely curious

In many cases you can actually insert outside numbers in the middle of the solver process via callbacks. For example see IncumbentUpdater at https://python-mip.readthedocs.io/en/latest/classes.html

And various C APIs for solvers have other callbacks

It's generally quite limited of course, for the reasons you mentioned.

I've used python-mip extensively, though only on problems with up to ~10k variables, and only using basic LP features (no advanced solver callbacks).

python-mip is substantially faster than cvxpy, pulp, and pyomo for constructing _and solving_ a very simple LP with 10k variables. It's on par even with cylp using vectorisation (maybe I haven't used cylp as well as I could (either way cylp is dreadful to use)). I say _and solving_ because for my use case it's the time to construct and solve relatively simple & small problems (like, <100ms overall) that I'm interested in, which includes anything like serialising the problem to a file and invoking the solver.

But keep in mind these kinds of benchmarks can depend a lot on how you're using the library. Like, are you spending almost all time actually solving the model? Then it probably doesn't matter what library you use to construct the model.

python-mip in general is pretty nice. It supports advanced solver features if you're into that sorta stuff. But their typing is a bit whacky (no py.typed file, use numbers.Real for typing which makes type checking a bit whacky). And it relies on a funky pre-compiled CBC binary (I had to compile it from scratch at a very specific commit to get it working on M1). python-mip creates/modifies variables in the underlying solver library (CBC/Gurobi) as you create/modify variables through the python interface (I believe this is an uncommon approach, but good). The internal data model of variables & linear expressions is quite clean and tidy.

Overall would recommend python-mip if you just need to solve MILP problems.

From my scrappy benchmark constructing and solving a very simple LP with 10k variables:

'benchmark_python_mip' took: 40.33 ms

'benchmark_cvxpy' took: 738.11 ms

'benchmark_pulp' took: 152.68 ms

'benchmark_pyomo' took: 191.66 ms

'benchmark_cylp' took: 38.91 ms

(this was a joke)