As AI agents take over coding, writing, debugging, and increasingly architectural decisions, engineers may seem close to obsolete. The argument here is that humans still make a measurable difference: if people added no value, engineers using similar models and tools should produce much more similar results. Instead, strong engineers continue to outperform average ones, with AI amplifying rather than erasing differences in skill. That human contribution appears in the ability to clarify an ill-defined problem, judge when a solution is overcomplicated or too simple, and consider whether it will remain workable years later. Engineers also rely on intuitions they cannot always explain, much like the shared sense that certain AI-generated writing feels like “slop.”
To explain these strengths, the piece connects engineering judgment to emergence and tacit knowledge. Emergence describes qualities arising from interactions among many smaller elements; tacit knowledge captures what people know without being able to put it fully into words. A person’s experience - conversations, mistakes, past work, fears, and ambitions - accumulates into instincts that shape what problems they notice and what solutions they trust. Models may surpass people in many abilities and learn patterns that resemble taste, but absorbing information is not the same as living through experiences. The conclusion is that engineers’ present leverage lies not only in explicit knowledge, but in the lived experience behind their judgment. Agents may eventually develop experiences and intuitions of their own, but until then, human context remains essential to directing work and deciding what is worth building.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Researchers tested 16 LLMs on all 55 TIS-100 puzzles, giving each one attempt and time to refine its code. The models found 133 new performance tradeoffs across 18 puzzles and set three outright cycle records, though human players improved on most solutions within days. (tis-100.wincon.dev)
Australia is considering a new criminal offence for AI companies that fail to take reasonable steps to prevent harm caused by their models. The government is also weighing an AI safety watchdog with power to impose significant fines, following a hack of a Medicare statistics portal involving OpenAI. (abc.net.au)
AI is increasingly becoming a gateway through which people find information. Audiences still resist having AI produce their news. (niemanlab.org)
Nvidia holds a central position in the AI industry, making disruption to the company potentially catastrophic for the wider ecosystem. The race for advanced AI is also intensifying economic and geopolitical conflict, while concentrated corporate power and infrastructure face growing political resistance. (france24.com)
LLM inference can require reading about 26 GB of model weights to generate each token from a 13-billion-parameter model. The explanation shows how model size drives memory bandwidth demands during text generation. (youtube.com)
Deletion Tuning matches full fine-tuning across three LLM post-training methods by zeroing fewer than 0.05% of model weights. Its compact masks also preserve prior-task accuracy better than full fine-tuning, LoRA, and random sparse tuning. (arxiv.org)
Today's best Hacker News stories, summarized and screenshotted, one email a day.