> Code becomes only one representation of thought among many overlapping ones.
This is wrong, code is the concrete "truth" being executed, the rest (plans, prompts, agent instructions) are just temporary artifacts used to generate the code. What's left is the code alone.
LLMs don't have any semantics, they can't execute anything with 100% certainty. So far programming languages are the only langugues that can do that.
That's technically correct but consider the following:
You're implementing quicksort. The actual algorithm is a conceptual piece of math. You write the code. But sometimes the code produces output that isn't sorted.
The code "wins" in that whatever the code says is what actually happens. But the code is wrong. It must be edited to match the algorithm. The _algorithm_ is the thing that is proven correct; the _algorithm_ is what we wanted to execute. The code is a representation of the algorithm.
Let’s say you were trying to say “I like bread” in French (which is “j’aime le pain”) but what come out is “j’aime le vin” (I like wine). It is the wrong statement, but it will produce an effect as the other party may bring you wine (which you may hate). The actual issue here isn’t that you didn’t speak French, but that you didn’t express what you wanted well.
So code is just a notation. But one that is executable. In your example, the code is working, it’s just not what you intended. The issue is not the code, the issue is that you don’t know code well enough to “speak” your algorithm.
And that is the main issue with a lot of LLM users. They are translating specs (which they do understand) to code (that they do not understand) and then relying on superficial outputs to see if they got it right. While also telling fluent speakers that they should follow the same workflow. At least with human languages, the other party may detect discrepancy and the fact that you’re not a native speaker. But a computer will execute things blindly.
No, I think it's correct. One artefact of programming for decades is that I ended up thinking in code directly without "translating" from English first.
Thus the push to use LLMs has felt a little bit like being a fluent French speaker being told that all of the best French writers are using google translate to write French translated from English now.
The necessary work is more than just the deliverable. Even a simple patch implies an indefinite number of other approaches not-taken, and sometimes not-taken for good reasons.
Sometimes we explicitly document those, sometimes we trust that a human expert decided against them. Those other aspects of work still exist, even if they are getting skipped/mismanaged with LLMs.
No? Code is literally instructions to the machine. Especially low level code. It's not a representation of anything, it's a list of commands in sequence for the computer to execute
The fact that you do not understand that is concerning.
I don't see the AI capacity jump in the recent months at all. For me it's more the opposite, CC works worse than a few months ago. Keeps forgetting the rules from CLAUDE.md, hallucinates function calls, generates tons of over-verbose plans, generates overengineered code. Where I find it a clear net-positive is pure frontend code (HTML + Tailwind), it's spaghetti but since it's just visualization, it's OK.
Hmm, your comment gave me the idea that maybe we should invent "What You Describe Is What You Get|. To replace HTML+Tailwind spaghetti with prompts generating it.
Sad to hear this attitude towards front-end code. Front-ends are so often already miswritten and full of accessibility pitfalls and I feel like LLMs are gonna dramatically magnify this problem :(
If you have a ChatGPT subscription, try Codex with GPT-5.2-High or 5.2-codex High? In my experience, while being much slower, it produces far better results than Opus and seems even more aggressively subsidized (more generous rate limits).
> However, at some point it seems that everyone at my stage is moving into management or higher level positions doing project management, meetings, architectural discussions
This impression is mostly generated by your head, to be more specific - by the social pressure to "make a career" and to regard technical skills as inferior to management skills. I don't think it's real, I saw many old people working as programmers, and there will be many more because there are more programmers among the current 20-30 year-olds, compared to the older generations. Also there's simply many more technical jobs available and there won't be enough management positions available to allow the switch for everyone. And the pandemic looks also like a quite big factor in reducing the number of management jobs (it looks like you often can work as usual without all the managers...).
The problem I experienced multiple times in 50-200 KLOC projects is not the time needed to import the modules, but the memory consumption caused by the imports. Moving some imports from top-level module statements to inner functions' code could improve the memory consumption several times, e.g. from 250MB per process to 80MB per process.
That's very nice code and Python is non-functional. I love its generators.
That's my version to make it more "functional":
from itertools import cycle, islice
from operator import add
fizz = cycle(['', '', 'Fizz'])
buzz = cycle(['', '', '', '', 'Buzz'])
strings = map(add, fizz, buzz)
strings_or_nos = (s or i for i, s in enumerate(strings, start=1))
print(list(islice(strings_or_nos, 100)))
Yes, this also seems weird to me. And I think it might be a cultural thing, I noticed that in Europe renting dedicated servers is far more popular than in the US.
IDK where you get the 'wages are a lot lower in Europe' thing. There are lower cost and higher cost places, but people are expensive and talent is mobile.
I don’t know, in the Bay Area 150k is entry level and big companies pay 250k or much higher to senior engineers. Whenever I hear about pay in Europe it’s a fraction of that. It may be worth it for a different lifestyle, I’m not making that argument. Regardless, from the company POV it must lead to different choices about buy vs. build.
Bay area or NYC sure, that's like being in London, but what about the Midwest, or any of the places where you can easily set up an IT shop without paying a premium for space?
- Get your own bedroom.
- 30 minute commute.
- Not everyone you meet will be in tech.
- You'll be rich compared to everyone else in town.
- Partner with local academic institutions to offer recognized research projects and training. 20% time or 3 month project stints etc. Adjunct professor ships for people with an existing research record.
- Throw in X free flights to east/west coast, home town etc.
there’s 30 minute commutes in san diego, la and sf if you live in the right areas. what about weather? cali has really nice weather. schools / doctoral candidates / etc are basically useless in the real world. if i’m making enough money the free flights are useless. nothing you said here makes me want to move to the middle of the US. exactly what i’m looking at is the lack of opportunity. say i decide to move for some job and hate it. now what? pick up the family and move again?
I ride a bicycle daily and vape e-cigs. If anything I notice an increased performance immediately after vaping, I feel like I'm less tired and have more energy.
Overall this is an interesting topic because contrary to popular opinion many athletes use nicotine and find that it increases their performance: Nicotine [...] use is widespread amongst professional team/strength sports (e.g. American football, ice hockey, wrestling, bobsleigh, gymnastics, rugby, skiing) whereby active consumption of nicotine and nicotine-containing substances in-competition occurs in approximately 25–50% of such athletes [1].
The meta-analysis [1] found that two studies observed an ergogenic effect, one an ergolytic with the remaining seven reporting no change.
I cycle daily and vape. I feel when I vape more, my performance on a bicycle deteriorates very noticeably. My effective lung volume is noticeably lower.
Those athletes have trained for years and have a heart that is well conditioned with strength and cardio. A normal person who is moderately physically active is nowhere near their fitness level. That’s why nicotine is a dangerous thing to gamble with. You might be fine, or you might have afib/vfib that leads to an MI. Rather avoid the risk. Plenty of more potent PEDs out there that don’t cause adverse cardiac events.
But even for fully on-site teams the preferred format of a "knowledge base" is digital (wiki, Google Docs, etc.), because it's something that lasts, can be improved over time and is always accessible. And having the discussions stored digitally (Github issues, emails) is also usually preferred because it's "asynchronous" and it's something that can be looked up in the future.
So on-site teams also use Wikis/issue trackers/emails a lot. The difference is that instead of a face-to-face chat remote teams have to use Slack. Which is also largely used in on-site teams... So my point is that some of the modern IT offices already resemble a "co-working" space because all the tools are already digital, and all meaningful knowledge sharing already happens in the digital space.
This is wrong, code is the concrete "truth" being executed, the rest (plans, prompts, agent instructions) are just temporary artifacts used to generate the code. What's left is the code alone.
LLMs don't have any semantics, they can't execute anything with 100% certainty. So far programming languages are the only langugues that can do that.