Go-attention: A full attention mechanism and transformer in pure Go

Re:Go-attention: A full attention mechanism and transformer ... by pizzafeelsright on Monday March 03, 2025 @02:18PM (#43245571)

I long for your confidence in your own ingenuity. If you think private repos on third party systems are hidden from training I suppose you have more trust in expendable developers working to maximize training.

At this point I am fairly convinced that software development is solved. The reality has not been understood by most as typing syntax is now a trivial task best left to AI.

[ Parent | on HN ]

Re:Go-attention: A full attention mechanism and transformer ... by sph on Monday March 03, 2025 @02:33PM (#43245744)

I have been migrating my private repos to a private Gitea instance of my own. I am well aware that anything on Github is used to train Copilot.

> At this point I am fairly convinced that software development is solved.

Writing code is 20% of software development, which is why I am still in demand even if I refuse to use LLM software. But employers and recruiters are not rational, and will soon rather hire cheap, expendable "monkeys on a typewriter" than experienced engineers.

[ Parent | on HN ]

Re:Go-attention: A full attention mechanism and transformer ... by imtringued on Monday March 03, 2025 @03:10PM (#43246091)

I wasted at least an hour today on waiting to anti virus software to let my software start.

No amount of AI will overcome the slowness of antivirus software.

Also, about software development being "solved". I beg to differ.

Sakana AI failed to reliably produce working CUDA kernels and that was in a domain where the task was explicitly specified as pytorch code, so it was, in a way, just a glorified syntax typing demonstration.

[ Parent | on HN ]