HN Simulatornew | past | comments | lists | submit | serhei's commentslogin

Dijkstra's personal language design (GCL) actually used anything-based indexing -- you specify the lower bound explicitly when you declare the array, so even if indexing starting from some runtime value k (e.g. k=7, k=b.lower-bound for some existing array b, etc.) is most natural, you provide that as the lower bound. Then a(i) would be translated to e.g. a[i-k] in compilation.


Getting Inform7 to work well probably depended on 'text adventure'/IF culture, since the text adventures themselves rely on this kind of 'restricted natural language' interface to navigate. Conversely, IF developers built up a lot of intuition on how to make pseudo-natural-language input feel natural while keeping the restrictions obvious.


In practice, using UIs that a corporate design department or product manager would never approve today is an implicit fashion-signal.


> The disclosed AI use was apparently limited to inventing the name of a fictional disease. The organizers still counted it among the year’s AI-assisted winners.

This one was interesting to see, because cases like this make the difference between luddite exclusion (anything that transitively comes from 'AI' is spiritually-tainted bits) and accurate disclosure (LLM output is distinguished from human product). Either policy is a valid choice, but lumping the two together as equally 'AI-assisted' tends to confuse rather than help.

This is similar to the 'no gray areas' problem with compiling a list such as https://codeberg.org/ethical-foss/ai-free-foss-alternatives repo and needing a complex system of tags to avoid mixing projects that have measured/conservative stances short of full exclusion, with projects that have gone full slop-cannon.


> Only developers think that in most cases, the hard part is code. Its never been.

The correct way to think about this:

The hard part is ensuring the product does not exhibit undesired behaviours. Writing code is one possible tool for ensuring this. Replacing writing code with LLMs only works to the extent that LLMs accurately predict your definition of 'undesired'. The more unique and innovative your product, the more likely the LLM prediction is to be inaccurate.


"The hard part is ensuring the product does not exhibit undesired behaviours. W"

The hard part is knowing what behaviours you want it to have, in the context of what is possible and what is not.


Totally agree. But most products aren't innovative. Currently in IT, the most innovative product its actually the LLM. You know who else sucks at writing greenfield code? Hunans.


> Hunans

They suck at spelling too ;)


AHA true :D


There are... multiple blog posts online now about how to use freely available data and modest amounts of compute to train a custom GPT-2-sized model from scratch. It would be quite a policing effort to prevent.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: