Parsing the universe as Pancomputationalist. Dad, the good guy, unbound Curiosity. Solving easy, hard, extremely hard problems using Computers. UG at IISc '16.
When you see the doctors coming out in hordes to confidently argue against new scanning technologies powered by advanced math and statistics, just remember that the vast majority of them are borderline innumerate and can’t even understand what p-values are:
TFR is <=1.5 in states with 581 million Indians. I argue that since many families will opt to be child free or have one kid, we should support those who want large families. TFR is currently dropping among all sub groups, those who use public schools and hospitals, those who use private facilities, rural and urban areas, etc.
@wshxnv@thsottiaux Bro take it easy people are using these things with claude, grok, perplexity and everything for months. It's just a simple MCP connector. Coding does increase the utility but it will also cause load to servers which are already in scarcity.
To put it simply, GUI doesn't expose the full potential of computers and computing. It's like riding a superbike but with bicycle paddles.
Out of last 22 years of my computer use, atleast 18 years of that is only CLI and minimal mouse and clicks for browsers. I'm happy browsers being replaced as intermediary to information by AI enabled tooling. Once in a while I do have to use windows in VM for one of tasks.
Seriously very useful tech. I’ve experimented with these ideas as a hobby with friends as a Professor at IISc was into these knee deep inspiring us, almost a decade ago!
Converting natural-language corpora into formal claims / typed ASTs / FOL or propositional representations, then checking them with deterministic systems like SMT solvers, Z3, Prolog, etc.
In my experiments I have found you can to a large extent make things formally verifiable, but yet a lot of probabilistic elements still remains and I think this is mathematically impossible to reduce, we proved it, I think using Turing machine/Church encoding like constructs, though don't quote me on that. There is no point pretending every domain can be fully reduced to deterministic proof.
@jxnlco It would be interesting to have it properly, right now I believe Codex is overfit for gpt models, so glm doesn't work upto it's full potential, it stops too early.