barrkel

There comes a point in a startup's life where more controls are needed. Red tape. The stuff that slows down the big boys. Problem is, the red tape is scar tissue from previous informal process failures.

I think these control rooms were superior in some respects to modern software system observability.

- modelling the system rather than implementation (system status rather than many individual service statuses)

- supporting causal reasoning: the control flow on top means you can trace failure modes back, visually; software systems typically only model their own ontology, and you need to look somewhere else for the next abstraction down

- surface state first rather than time series; a pretty graph is nice to look at, but for actionability sometimes what you need is the flashing red light

- prioritize first-out indicator. In a complex system with lots of alerts, the most important diagnostic alert is often the first one - the rest are downstream and contribute to alert fatigue, despite them probably being more important business metrics

Is it being used in production anywhere, even if only a toy app?

I know you say it's not production ready and not optimized yet, but in the same breath - in your comment here - you say it's already faster.

My experience has been escalating fragmentation past 90%, along with lower performance owing to the fragmentation.

It's not actually. Think about it for a moment and you'll see life is full of tensions between preventing crimes and freedom / utility.

It's most obvious on the roads. Few non-commercial vehicles will limit your speed to the national maximum. Wouldn't a strict interpretation of your opinion imply speed governors?

I didn't put them on the same level.

At the same time, one should acknowledge that not all tasks are on the same level.

People are stochastic. You build reliable processes out of unreliable parts with feedback and self-correcting mechanisms. AI is not actually magically special in this regard. It has higher variance and we're still figuring out how to get all the tradeoffs right.

Repeal Town and Country Planning Act 1947

You see a lot of hobby shops in ultra-wealthy areas of major metropolises. Tiny art studios, interior decorators with a handful of items in stock, boutique fashion shops.

Can you try and tune your Claude or whatever LLM you're using for your text to phrase things in plain English. Way less use of antithesis, at least. You can probably find a skill for it, if not get an LLM to write your own.

I never want to read the phrase "heavy lifting" again.

The second mode is another way of doing the first, actually.

Either way, working people's output today needs to support the non-working people today (unless you let them starve). Whether that is done via taxation or via wealth accumulation, one way or another, more and more young people will be supporting old people.

And it's not even clear the wealth accumulation angle is good. It requires young people to own a lower amount of wealth so that old people can dangle it in from of them to keep them supported.

I found it interesting that vLLM was dismissed as slower than llama.cpp.

IME vLLM is quite a bit faster than llama.cpp but where it really wipes the floor with it is in batching concurrent load. The downside is that it is dramatically less flexible in terms of tweaking. It gives you very few options for running quantized weights. It takes a lot longer to start up because it optimizes the compute graph. So for single user experimentation on a model that's a bit too big for your box, vLLM is just going to be frustrating.

That revenue exceeds cost of revenue so tokens aren't being subsidized...

Do you think token completion endpoints are the final form for AI APIs?

It takes time to re-ground yourself, to renew your ambitions to new circumstances.

It usually takes more than a year even for employees.

You don't need to work, but you need to have value in your tribe. You need the respect of peers. You need to feel part of a team. You need to feel like you contribute something to the lives of those you love.

Working to keep a roof over the head of yourself and those you love is an identity. It's social proof that you have value, that you can do something for someone else.

It's my understanding that a big part of WSL1 performance loss comes from the relatively thick layered filesystem architecture on Windows.

Since git and nodejs are both common in modern development and are expected to work efficiently with huge numbers of files, this was a real bottleneck and it couldn't easily be tackled without threatening backward compatibility.

The trouble is thinking in terms of class.

It encourages one to generalize from instances to the group, and from the group to individuals.

Cherry-pick your examples and you smear the group. Invoke the smeared group, and you target the individual.

It's structurally isomorphic to racism and any other -ism.

It only goes into the factory owner's pocket to the degree that the factory has no competitors (has pricing power) and the factory owners don't work for the factory (i.e. RSUs and the like).

Marx is great at building a narrative that generates resentment if you buy his frame. But you don't need to buy his frame, and if you don't, you suffer a lot less resentment. It's no way to live.

You had to deal with two flavors of pointer, near and far. Far pointers came with segment selector, for accessing more than 64k. Your choice of memory model influenced the defaults. You might use near pointers for internal references in a module, and far pointers for external references.

Rear view cameras are not a replacement for rear view mirrors, is my point. You use the two for different things. A rear view camera trying to replace a mirror would need a cleaning and wiping solution, something that we use rear window wipers for with analogue mirrors.

Containment of the execution environment isn't really the issue. It's API tokens that were designed with coarse permission scoping so agents get more power than they need. The risk isn't that your machine gets hacked. It's that your email gets deleted, or forwarded to someone who uses it to break into your other accounts via password recovery.

A statement that some proposition is true or false is usually less useful than a new framework for understanding the class of problem.

A machine that takes longer and longer to prove propositions in ever more inscrutable ways is hardly useful at all.

The machine too needs to produce more generalizable and comprehensible systems, for it to scale up its own conceptualization. Needing to load all the new mathematics in the context window won't be great either.

Rear view mirrors are needed all the time. We're not talking about reversing cameras here.

Quietly. Clean. Honest. Sharp take.

I learned about machine code and two's complement from the Machine Code for Beginners book. It gave me a head start in college 10 years later. With it, I got my Amstrad CPC 464 to run a loop maybe 100 faster than BASIC, and I was enlightened.

The Usborne books were the single strongest teacher I had getting into coding. I never owned any; I relied on the library.

In the earliest days, I didn't have a computer either. I'd read an Usborne book, then hang around computer stores poking at the 4+ years out of date Commodore 64s and CPC 464s and even the Acorn Archimedes (fanciest most capable BASIC), putting what I learned into practice. I'd even practice on VTech devices with two monochrome LCD lines of text, in toy shops, to get my fix.

I don't think I'd be where I am today without those books.

It's because Japanese society frowns on entrepreneurism, which means innovation needs to be attached to an existing company.

Immature kids - and kids are by definition immature - make dumb decisions.

Certainly there needs to be repercussions, the kids need to learn, and need to be warned in advance, but kids will do dumb things.

You can talk about news coverage, but you can't force 12 year olds to watch the news, or understand how their actions have consequences. When the consequences do come, the focus needs to be primary on rehabilitation and restitution for the first instance.