while im not necessarily a doomsayer about a recession, im not sure any of these are adequate buttresses against one
the s&p doesnt tend to go down until a recession is either imminent or happening, and modern economic policy may even prevent that due to the vastly expanded wealth disparity
the unemployment rate isn't really the tool it once was--if you include two groups that count as "employed" by the US unemployment rate--people seeking full time jobs but working only 1-34 hours a week, and people earning below the poverty line (under 26k pre-tax anually), the functional unemployment rate climbs to 24.9%--higher even than any month in the post-covid period of 2022-2024 (inclusive).
yes, more small businesses than ever, but growth last year was almost non-existent. and it generally proxies vs able-bodied adults, which would have seen an increase larger than the businesses.
gdp is high but much of that is ai company shuffle and debt-to-gdp ratio is rising once again--now the highest it's been since the 2020 massive covid bump
birth rates are broadly down as well, likely stemming from cost of living, and birth rates do tend to predict recessions.
it's fine to point at numbers but they don't really mean anything in isolation
the whole article kinda reads like "i have a leak in the basement of my house in the pacific northwest. the solution? im moving to nevada"
i dont dislike rust at all (infact, its rustler interop with elixir/erlang is great), but the article reframing a bunch of intentionl design choices (that i would broadly argue as good design choices) in golang as shortcomings is so weird (gc, generics, error handling, etc.). especially so when they're framed in such a way to make error-prone go seem inevitable, or are directly comparing well-written rust and poorly-written go. take, for example, the section on data races. the article broadly classifies rust as data race free and golang as full of synchronization issues. and this is true if you only actually care about data races (not race conditions broadly), assume all of your rust is safe rust, and none of your golang uses any of the available solutions (atomics, synchronization primitives, channels, etc.) to data races. yes, go leaves much of the behaviour up to the programmer. this isn't a downside.
more egregiously, the article glosses over two of the biggest and, to me, most critical differences between the two language. first, go compiles FAST. i can write something and test it immediately, including stepping through the code. i dont need to context switch away from the task and can easily, and quickly, program fixes and changes and features. this is such a huge development gain that switching away from it would require an incredibly good reason. secondly, the package structure of rust offers a clear vector for supply-chain attacks. not that golang is perfect in that sense, but it has a ton of factors that reduce the likelihood, and if i'm being really picky about safety it's going to be a big consideration.
I use both languages and in spirit you are right, btu in my experience its more nuanced. first of all the way inline unittests in the same file is a way to have very fast cycle times in TDD too in many situations, still slower than go full compile yet much less painful than full recompile in rust. second, you typically need way less debugging cycles in rust to begin with. so its more like slower but fewer cycles.
I think you've misread the article, specifically the purpose of the Brooks quotes. They're clearly not to denigrate anyone for wanting a more useful or convenient way of generating software--they're specifically about the "LLMs will obliterate software engineering as a profession" claims put forth by many LLM marketers. In fact, Brooks is never once mentioned in the "Power to the people?" section of the essay.
Generally, the whole point of the "Power to the people?" (and to some extent the "On being left behind") section(s) is to underscore the two antithetical claims made by many LLM marketers:
1. LLMs are so powerful and so natural and easy that someone with no experience can create amazing software, and
2. LLM usage is a core skill, one that if you don't begin training now you'll be left behind.
Obviously, both of these can't be simultaneously 100% true--either it's easy enough for the non-programming layperson to successfully generate software for an intentional purpose, or, LLM assisted programming is a skill you need to train to avoid professional obsolescence in modern society. So, the article disagrees with the majority of both claims, and accepts a weakened/minor portion of each:
1. LLM output is easy to generate but accurate prompting matters, and
2. when used for software development professionally, some amount of skilled human intervention does indeed seem necessary.
And now these two claims do align.
However, if professional software engineers who work with and read code constantly, armed with the best software practices to aid LLMs we can determine, cannot use modern AI tools without shooting their feet off at relatively frequent rates, certainly you'd expect the layperson who must put an even greater amount of undue faith in the validity of the results to be at extremely high-risk of foot-shooting. It's not "gatekeeping" to forewarn people against unwarranted trust in LLM output, nor is it "gatekeeping" to suggest that modern tech communicators/marketers describing an overly flowery LLM tooling landscape might be doing people a disservice.
"dont prematurely optimize" is talking about optimization SEPERATE from architecture (and how pre-mature optimization as Knuth describes gets in the way of correct architecture). nowadays, almost any optimization purely distinct from architecture is handled either by cpu arch/logic improvements, compiler/interpreter improvements, or buried in the overwhelming tide of cpu speedups, so TODAY these terms get conflated as architecture IS typically how we optimize.
Weird take--SOLID, to me (I work in embedded but have done basically everything), represents a system of design principles that mean well and are probably fine in a heavily OO environment 80% of the time but resoundingly end up prime examples of the pareto principle.
maybe, but they're burning the compute regardless. it seems ostensibly likely that reducing the ROI for compute burnt will cause less compute to be burnt long term
Elemental sodium is reactive. Ionic sodium is not, lest you blow up your dinner. Furthermore, the lithium part of a Li-ion battery isn't the flammable part, the electrolyte is.
> If you want to replace FF there is exactly one solution, that's nuclear.
> Stop acting like you care about this issue. You have never cared enough to learn about it, so until you do, stop spreading misinformation about how physics works.
It's wild for you, in particular, to take such a weirdly aggressive stance here. Zero basis in reality, just virtue signaling.
> The FDA is partially to blame for this situation: ...
> The cost of performing a New Drug Application starts in the mid hundreds of millions of dollars range and can extend into the billions for some drugs.
> So nobody could feasibly introduce it to the market here without investing $500 million or more up front. At that price, your only viable option is to stick a big price tag on it and try to milk that money back from insurers.
It's interesting that you seem so passionate about this because you're totally incorrect. The cost of a NDA for a novel prescription drug requiring clinical data (the most expensive application) is ~$4.5mil. In fact, the estimated TOTAL revenue to the FDA from ALL PD application fees in FY 2025 is ~$1.3billion (or, just under 300 novel prescription drugs). So, obviously, FDA fees can't be as much as you're claiming.
What you're actually describing is the total cost of the entire drug development pipeline (research, design, lab costs, chemical costs, application costs, marketing costs, etc.) to develop a brand new, novel drug. And it's only ~$200m, increasing to $500m if you include dead ends / failures in the process, and ~$900m if you include both failures and capital costs--yep, that's right the capital costs alone are almost as much as the entire rest of the drug development pipeline.
any commercial rtos shop where QNX may be appropriate is either using 1. some wacky expensive proprietary rtos that you've never heard of, 2. freertos or 3. real-time linux depending on what they need. asking what makes QNX a compelling rtos when freertos exists, is widely supported and used, and has an MIT license is a very valid question.
further, no one in embedded actually cares what RTOS you used. they are all similar enough that you won't get stuck if it's a brand new RTOS
QNX is heavily used in industries where functional safety or particular high assurance models are required.
Sure FreeRTOS has a SafeRTOS mode, but its not sufficiently functional for a modern ADAS stack or complex robotics systems. QNX is used in all major automotive companies around the world for a reason, and a crucial part of NVIDIA's DriveOS stack.
QNX is in a space with few competitors. FreeRTOS or ThreadX are designed to provide microcontrollers with scheduling and memory management functionality. They don't depend on fancy things like MMUs or provide frameworks for networking or file systems out for the box. The flipside is that you can compile them down to maybe 30kB of machine code.
QNX is designed for more powerful and featureful hardware to drive a software stack with true process isolation and generally provide the bells and whistles of general purpose OS on top of a hard realtime core. It can run complex GUIs without sacrificing its real time capabilities. Not many competitors live in that particular space.
Without Oxford comma: "We invited JFK, the stripper and Stalin." [three distinct items in the list]
With Oxford comma: "We invited JFK, the stripper, and Stalin."[two named items in the list with an appositive affirming that we're talking about JFK the stripper and not the former president]
Yes, changing the order and the number of items (plural strippers in original to singular stripper in yours) changes the meaning. That is unsurprising.
the s&p doesnt tend to go down until a recession is either imminent or happening, and modern economic policy may even prevent that due to the vastly expanded wealth disparity
the unemployment rate isn't really the tool it once was--if you include two groups that count as "employed" by the US unemployment rate--people seeking full time jobs but working only 1-34 hours a week, and people earning below the poverty line (under 26k pre-tax anually), the functional unemployment rate climbs to 24.9%--higher even than any month in the post-covid period of 2022-2024 (inclusive).
yes, more small businesses than ever, but growth last year was almost non-existent. and it generally proxies vs able-bodied adults, which would have seen an increase larger than the businesses.
gdp is high but much of that is ai company shuffle and debt-to-gdp ratio is rising once again--now the highest it's been since the 2020 massive covid bump
birth rates are broadly down as well, likely stemming from cost of living, and birth rates do tend to predict recessions.
it's fine to point at numbers but they don't really mean anything in isolation
reply