@ghostof142@thdxr Lol clearly someone who loves reading ancient writings would find time to reply to this tweet nearly a year later. In all fairness It's been a brutal year for Bitcoin. Fingers crossed there are better days to come though.
@antoniosarosi I miss the world where not everything a software engineer did was evaluated by the raw value generated. Only because the best way to improve is often to build the harder tool, with the least helpful software, but do lots of exploring of bad ideas and concepts that are wrong.
@fjzeit Maybe I'm old fashioned but I miss people reviewing code written by people. There was a sense of pride involved in the process and acknowledgement of each others ingenuity and creativity. Agents deliver sheer value but lose so many other things along the way.
@Mike_Preston17 Yea the problem is so many people who only half ass code some generic web apps to begin with, think this is the greatest thing of all time. They aren't building airline software or life saving medical devices or they would be thinking very differently about reading the code.
@michael_kove@BigWhiz_001 It is an extreme example, but it points to a necessary requirement for certain types of software, that will require a level of scrutiny that is beyond just a bunch of unit tests that are generated by the AI. You must understand the problem to a degree that makes the AI useless.
@michael_kove I work on medical devices, but similar to aerospace, I couldn't imagine not knowing how every single byte of memory is being allocated, let alone just trusting an AI to completely drive the development process for me without ever reading the code it is generating.
@svpino So basically we shouldn't be flexing about writing code that runs aerospace or life saving medical equipment... We should just let the models do that instead and stop reading the code it generates??
@BDurhamSports@Matty_KCSN Look I am not even an Allen hater by any means. I just can't see any argument for Allen being ahead of Mahomes in a historical sense. I am fine with saying he has been better the past few years but for the decade it is largely owned by Mahomes still.
@BDurhamSports@Matty_KCSN Mahomes has beaten him 4/4 times in the post season, every post season stat for Mahomes shows a wider gap than every regular season stat is for Allen over Mahomes.
I feel like we're all part of a massive experiment. At first AI blew my mind at what it could do for SWE. Now I'm struggling to understand what benefits it really brings. Software engineering just doesn't seem fun anymore, I'm not even sure AI is more efficient in the long run.
We're just all using it because we still believe and FOMO. I hate reviewing AI code, so then I just don't look at it anymore, I test the work extensively, I'm just a QA at this point.
So what's the point. I feel like in the future Engineers that can actually write code will be able to command huge salaries when everyone else is just pumping out AI slop.
@camsoft2000 This is the problem with fully leaning into agents. You aren't producing better code than you would have before AI (just more). There is a balanced approach to prompting that actually makes better code but requires people to do >75% of coding by hand. It's not faster though
@LewisCTech This is obvious. They all have incentives that align with AI. However it does feel like AI is getting harder to ignore as a useful tool. I still keep mine in the browser as much as possible though.
@soci9talm9naci@rutherfordml@chinedu_10 Yea I never said it wouldn't write better code than a majority of humans, I just meant at least 1 single human on this planet has to know better than it in some regard for it to pass the knowledge to the LLM for its training to be built.
@soci9talm9naci@rutherfordml@chinedu_10 Yea that's called reinforcement learning. A technique that isn't used for training LLMs and if it was it would produce worse results than the way LLMs are currently trained.
@rutherfordml@chinedu_10 This is true but another term for those patterns is called hallucination. I am not saying it can't come up with new patterns but its only able to know which patterns are valid based on human ingenuity through feedback, not its own.
@rutherfordml@chinedu_10 How does a model trained on humans become better than the humans who trained it? At least 1 person on the planet must be smarter than the model to pass along the wisdom that makes it exceptional at programming.
@MarcJSchmidt This is just an obvious fact. I know people compare AI to historical technologies, but there is a significant difference because the intelligence of the human was never the thing being replaced. Not saying it's good or bad, just very different.