Welcome to the main community in programming.dev! Feel free to post anything relating to programming here!
Cross posting is strongly encouraged in the instance. If you feel your post or another person's post makes sense in another community cross post into it.
Hope you enjoy the instance!
Rules
Follow the wormhole through a path of communities !webdev@programming.dev
Using LLMs to write code was never about making it easier. Its only about making it faster. All product teams only care about how fast you can get a product to market. Even before LLMs, that's all they cared about. That's why we have to sit through excruciating meetings playing planning poker, only to constantly remind them that two story points does not mean two days.
Edit: I knew I was going to trigger everyone by mentioning story points
Story points are such entire bullshit.
How do you measure "effort"? What is a unit of effort? How do I know whether a task is a 2 or a 3 efforts? What is the actual difference there? "It's the complexity, the work load, and the risks and uncertainties". Ok... but are those things truly measurable? Are they equally weighted? What complexity takes it from a 2 to a 3? Are we just going off of vibes here? Does a 2 to me mean the same thing as a 3 to you?
Why would effort scale with a fibonacci scale? How can non subject matter experts estimate effort on a task they aren't familiar with from a Jira Card Description? Do you measure effort of dev work and qa work? Or do you separate those things? If dev work is 3 story points and qa work is 3 story points, what is the combined effort on the fibonacci scale since there is no 6? Or is it just a 3 between them because the dev and qa work are assumed for all tickets?
But also, "It's not about time, it's about effort". OK then why are we measuring velocity in story points per 2 week sprint. And we have a story point capacity per two week sprint. And we adjust that capacity directly against the number of days people are taking off this sprint. So it's absolutely, positively, not a measure of the amount of time needed. But we do allot story points relative to manhours available.
sigh. I wonder if I'd rather just have bullshit deadlines then play another pointless game of planning poker with only half of my team barely participating and the scrum master trying to get enthusiastic consensus on a scale that is, at best, about how you feel about a task you are probably just learning about and don't fully understand.
So much of agile "process" gets so close to the point while still missing it, it's infuriating. Absolutely 100% agree that story points are mostly meaningless, but story pointing shouldn't be the point of backlog refinement. The point should be to discuss upcoming work, clarify things, ask questions, make sure everyone's on the same page, moving in the same direction, and making sure all the work is accounted for. Story points can be a nice punctuation to all that discussion and planning effort, but for the most part it's nothing more than a contrivance to satisfy management's boner for tracking useless metrics.
I totally agree about story points being bollocks. Their supposed justification is that you can measure velocity by seeing how many story points you do per month, but that only works if there's no feedback from velocity to story point estimation which is obviously not the case.
So in practice they just become a stupid proxy for time. "1 story point is an hour right?"
The Fibonacci thing makes some sense to me though - it is a way of suggesting that bigger estimates are less accurate.
Of course it would be much better if you could actually put an accuracy directly into Jira or whatever (e.g. 10-30 hours), but that would apparently be way too complicated. You could also get that data fairly nicely from planning poker (which I also think is pretty good, but you throw away all the information about uncertainty that you just spent ages discovering!)
LLMs also enables managers and businesses who want to make money with shitty products, which is probably a higher detriment to code quality than the low-level difficulties of programming ever was.
Either way, IMO the prime use of AI, including LLMs, is mass surveillance. And no, that's not a good thing.
i find it funny how people freak out over Cambridge analytica, but when you change the name to AI and increase the data 1000x from just stealing user data from facebook to user data on every service "everyone seems to love it" and feed it more data.
Agreed. I see some good use cases for the tech but the surveillance is obviously disproportionately bad.
Yeah for sure. Facial recognition as used in mass surveillance AFAIK uses very similar machine learning tech you can use quite successfully (AFAIK) to aid diagnostics in medicine, environmental conservationism etc.
Machine learning was being used for security/traffic cameras for years before LLMs or Flock so yeah. Detecting patterns/shapes in images and text is what this tech was created for.
How is LLM facilitating mass surveillance (except help build nasty data processing tools). Reading through unstructured ingested data?
Facial recognition, gait recognition etc. (i.e. a different kind of machine learning) is probably more important for surveillance, but LLMs can play a role, too. Websites, unencrypted messengers, email servers etc. have been collecting craptons of text data, but before LLMs it was fairly difficult to analyse all of it if you don't have a place to start like "user username with IP address xxx.yyy.zzz.pp called the president a butthole on dd/mm/yyyy". You could always filter for keywords, but people get creative about it, which is substantially harder when LLMs are used for analyzing. So yeah, "nasty data processing tools", and that's actually a big issue because of how nicely it slots into the network of mass data collection that companies have been building for targeted advertising (or so they say, to me it was always clear that a lot of them were already using it for more nefarious purposes).
And e.g. I assume that you can identify most people just by their writing style across tens of otherwise unrelated accounts.
Also, it's not like you can't use the hardware that's training or running LLMs for other kinds of machine learning.
I'm guessing they meant AI in general
It won't ever be easier except in unique cases or during the transitions. Companies will modify incentives and disincentives to encourage the same amount of effort from you but take any increased productivity for themselves, same as they always have and always wil. It's a consistent pattern of behavior that emerges from the incentives and disincentives of the capital system itself.
We will never be handed better lives. We have to take it from them to ever have it. I wish we'd all accepted this already and begin working together to make it happen.
but take any increased productivity for themselves
Many people seem confused and this.
Worker makes charts and reports. Owner sells for profit. Worker suddenly is making double the charts. Owner makes double the profit. Worker... ???
Just like... What do they expect to happen.
This classical model frequently positioned memory and recall as both the central enablers of, and bottlenecks in, software development.
Today’s AI-powered coding assistants are changing that. These tools function as external memory systems, offloading syntax recall, boilerplate generation, and API usage from human to machine memory.
This is a good way of putting it imo
This is why I only use AI as a last resort tool, I never use AI otherwise. AI can't really solve problems as well as humans do. I see it as only a tool, but last resort tool. I have only resorted to it 5x in the span of two years.
Optimocracy values reliable software over AI hype.
