Posted by poisonfountain 7 hours ago
The fact that the author can articulate _why_ the AI is getting so good is kind of a moat for specialist, right? Imagine a layman prompting without domain expertise:
"There is likely a race condition here + [long-winded explanation and analysis carefully guiding the AI]"
Degenerates to:
"This button is not working, please fix. I don't care about code. Decide yourself"
Degenerates to:
"Claude make me money"
How is that true? I've been using Opus on an industry scale over last 6 months and this is just not real.
It has consistently with a certain percentage of chance each time (and no claude.md and skills do not stop it fully):
* Suggested to remove tests to allow for things to pass
* Suggested remove an error so that things can be "unblocked"
* Suggested to use a second path when the original path ran into problem instead of making the original path accomodate for that possibility.
* Suggested or silently added "features" or "guardrail" that I don't want.
* Can be left unsupervised only if given a goal that it can verify against itself. Without such clear goal (e.g. this test in the integration environment must be fixed), it flounders.
I'm not using just the native harness (e.g. CC) either, with additional, customized harness, the behavior improves somewhat but are still fundamentally constrained and cannot really be trusted without verification.
See my methodology (100% handwritten): https://aperocky.com/blog/post.html?slug=agentic-development....
Being a heavy user I think I've ran into every single hallucination that the model can do over development release and operations. I am still a heavy user but there are a lot of value in recognizing where exactly LLM's limit is and work around that.
Developers are concerned about jobs going away, but how often are they pushing back in their orgs about how AI works? In response to "are you using AI to move faster," how many are responding with "yes, but there are some things you should know..."?
If there's no pushback and just pure acceptance of stuff like tokenmaxxing, then what does anybody expect when the broader narrative around AI is that it can help a novice to grind out miracles (i.e., "holy crap, if this is what a novice can do, what can an expert do?!")?
Of course leadership is confused because (it seems) few are asserting expertise, saying "no," and stating a clear case as to why they're doing that.
The default excuse is "I don't want to lose my job" (which is a fair reaction to all of this, especially these days), but it's worth considering when/how that choice is actually just shooting future you in the foot later. It seems there's a broader trend toward compliance more than there is "you hired me to do this job properly, did you not?"
However with AI, it feels different. I have seen both technical and non-technical managers tell engineers something to the effect of "you aren't prompting correctly" if they aren't able to get the task done within some preferred time frame.
We are seeing the industry revive metrics like lines of code, number of tickets closed, bug's found (looking at you Mythos), and now even "tokenmaxxing". It's exhausting to push back on. These are all things that we know will be gamed. But the individual that brings this up might be viewed as "anti-ai" or something.
If you're an IC, I do think the best thing to do is just go along with it. Sooner or later we will see more shocked-pikachu-faced executives when they realize that engineers are spending tokens just for the sake of it.
I personally think the best thing to do is start retraining now so you aren't screwed by the time this all topples
Current LLMs are still kind of shit at actually programming so many jobs do still care to have professional programmers. However, I think it's evident that if things stand where they are, employers will care to have far fewer of them, at least of highly paid highly experienced programmers. If this is the state we're in with LLM adoption when they can't help but create the same helper functions 15 times, god knows we're screwed.
So we should probably work on clearing out our debts and figuring out what else we might want to do with our time, I reckon.
I'm still going to try to do a good job. I'm still trying to learn the best effective ways to apply current LLMs (Right now I still prefer to mostly write code myself but have been using LLMs to bang code into shape via iterative code review; this is a way to exploit LLMs to make better code, especially applicable if your velocity was already good.)
It's the exact same story that we've heard countless times by now. Hosted on a blog with just a single post. Named in a way that suggests that said blog was created for this very single post.
What is there to learn from this other than LLMs seem to be bad for some people's psyches and that AI companies need these very stories to not get their funding shut down?
Would you put a "Hey i'm feeling a little useless" post on your main blog / linkedin?
It might be easier to adapt to this new tech when you're 19 compared to when you're 59.
But honestly, this discussion _also_ has happened ad-nauseam by now. Everything that was worth saying has been said. And then some.
People don't actually want to talk about LLMs. They want a hug. And that's fine, human and all.
But could you please just start asking for hugs instead of encoding that into vaguely profound sounding takes on AI? I'm tired of this play pretend.
Monopolies will continue as Token prices continue to rise.
Programming, logic, etc are skills and toolkits. The optimal state of society is everybody being able to apply them, not just the enlightened compsci caste. There was a time in the past where scribes were paid nice cash for their efforts, too.
I guess the lesson to learn here is treating a toolkit as an identity and job for life. By virturee of the essence of the job itself - if the tool gets cheaper and more widespread, it's aactually success, not betrayal.
Maybe using writing as an analogy is flawed, but most of humanity having 'writing' as a core skill did enable many other things, even if oral storytelling cultures suffered at its hand.
At its core, tech is all about breaking through inefficiencies and barriers. Does it matter if people can't code python if people demand government systems be frictionless in the year 2500?
The thing many people are ringing the alarms over is the offloading of critical thinking and knowledge work to LLMs.
I personally think the alarm ringers are mainly the privileged elite who are scared of their moats beyond filled in. LLMs have effectively broken down the gates of access to knowledge. In a diverse world, having more people being empowered to do more things has to be a net positive.
Once people get over a few hurdles, things like: >tech's too confusing >$20 is a lot of money to spend on a subscription >AI is just a fancy search engine >AI will do all the work for me
You start unlocking a fair bit of creativity in people. I mean, all this is brand new stuff even for tech-savvy people. It'll take a while for the genuinely useful uses to dissipate out into the maasses.
Not everything has to be a billion dollar business.
1) Train AI to replace human work. This gives you 50% quality for 10% cost. 2) Train AI to assist human workers. This gives you 200% quality for 110% cost.
Most companies will go with option 1, and it's a race to the bottom. Eventually, someone will go with option 2 and gather up all of the pieces and take over the market.
If you train an AI in one thing it will become better in the other.