Theme 1 – Skepticism about LLMs achieving true autonomous self‑improvement
- rmunn: “no, of course not, in fact they will never be capable of achieving good results with that technique.”
- rmunn: “they will end up training the LLMs on their own output and lead to the inability to distinguish reality from hallucination.”
- VCFundedGenYer: “Betteridge's Law. No. And it never will.”
Theme 2 – LLMs are useful for routine, low‑ambiguity tasks (monitoring, debugging, hyper‑parameter tweaks)
- janalsncm: “Claude can handle this. There is very little ambiguity, and we are basically just looking to maximize some metric under a set of constraints.”
- janalsncm: “If your training run dies at 1 am… you can lose up to 18 hours of work… LLMs are usually capable of … tweaking a single hyperparameter and rebooting.”
- janalsncm: “Even just that task means I can kick off multiple runs over the weekend and have confidence they’ll finish.”
Theme 3 – Human judgement remains essential for ambiguous or innovative work
- janalsncm: “Many business processes are not like that… it isn’t that easy to say whether a system has done a good job or not… LLMs can help with this a lot but they have bad judgement because it requires talking to people.”
- rmunn: “What you're describing could have been done with a short script… the LLM's being able to parse the error message… is a definite improvement … but I'd classify this as LLM being used to automate a sysadmin task, rather than calling that self‑training.”
- janalsncm: “recursive self improvement just means tools helping us to create better tools.”