Mark Bolusmjak @🏠

The fundamental goal of a programmer is not to get shit done. It's to understand, completely, what you've done. Without that, you're fucked.

Next we can tell construction workers that tools are heavy and expensive . Use your bare hands to hammer nails. Free and lightweight!

@VictorTaelin Strand, FLENG (Prologs). See call-with-current-continuation.org/articles/the-j… and call-with-current-continuation.org/software.html

Looking more examples of simple, self-contained (i.e., could be implemented as a small Haskell file) symbolic AI algorithms that perform well and are heavy on dynamic allocations, pattern-matching, closures and so on. Specially if there are published benchmarks. Ideas?

@nathanmarz The more expressive a language is (runnable programs per volume of code) the more easily it is to accidentally create incorrect runnable programs. DSLs give you expressivity where you need it while simultaneously enforcing correct semantics for the domain.

Why are programmers so frequently in such a rush to give up programmability? You so often see config files instead of code and limited DSLs instead of libraries.

@AgileJebrim That’s because I don’t program GPUs.

@z5h You don’t look like you program GPUs.

Fire invented. Majority: bah, it uses up too much wood. Wheel invented. Majority: ooh, can I get one in square please.

@VictorTaelin As Epictetus says. "Some things are in our control and others not." In your control are the points you laid out. Not in your control is misinterpretation. Don't worry, keep doing your good work. There will be no shortage of smart and industrious people who appreciate it.

Seems like: 1. Writing only truthful information 2. Triple-checking every claim 3. Spamming disclaimers for every possible misinterpretation Is still not enough to prevent people from misinterpreting you. What else I could I possibly have done here?

Curreent state of my desk.

@ysegmond I hope this works. I hope it's done right.

@skdh Any agenda other than science must necessarily be anti-science.

More attacks on scientists for doing their job, very worrisome nytimes.com/2024/04/23/wor…

@VictorTaelin @alittletyper But if you do full Prolog I’ll be your biggest fan.

@VictorTaelin @alittletyper Look to miniKanren instead. Pure logic, designed to be embedded and extended. Doable in one evening. Would its underlying lazy-stream approach parralelize?

simple algorithms you'd like us to try on HVM-CUDA ↓

@VictorTaelin Prolog DCGs, a dead simple idea once you see it, took a decade of work to figure out.

The cold truth about HVM2 is that my last 10 years of work could've been a single weekend project, if only I was 10% smarter. HVM-CUDA is a 1k-LOC file that could be written in two days, given the knowledge I have today. It is hilariously simple, and there is nothing else to it. This is kinda sad honestly, because I can actually imagine someone much smarter than me looking at the problem and immediately coming up with the right solution. I didn't take 10 years to write this code... I took that long to find the right abstractions - which were never complex, to begin with. 💀 To be fair, what made it work is something I always knew I could do. I just never tried it, because I assumed it wouldn't be fast. I just... constantly dismissed it. And, I mean, nobody else managed to make a fast GPU IC evaluator either, including researchers that worked in this problem for decades. So, it isn't like I'm the dumbest person alive either. Still... It is unfortunate to think I could've done it earlier, but it is comforting to know I have the solution now, and kinda vindicating to finally see it working as I always thought it would. It is a good bad feeling. I'd describe it as eerily soothing. I'm ok with this outcome.

Contradicting the law of noncontradiction. ?- \+ (compare(A,A,B), \+ compare(A,A,B)). false. #prolog

@dhh doesn’t appear to understand the difference between a logical argument and unexamined anecdotes. That’s probably why name-calling is used to sound more convincing.

DHH enjoys the freedom of having more ways to write incorrect software.

Imagine London at 38.9 C on Jan 1st in Winter This is just what happened in Chile and it did not even make main stream news twitter.com/extremetemps/s…

@aaronwhite xkcd.com/505/

@aaronwhite Given that the “Rule 110 automaton” can be used to run any LLM we have, and given that Rule 110 can be implemented by placing rocks on a large beach, it seems fair to ask if subjective experience lies in the placed rocks, the current rock, or in the action of placing the rock.

The hypothesis that LLMs *aren't* sentient is temporarily backwards... in a few years, kids will start growing up w/ AI friends, indistinguishable from other online friends. The burden of proof will shift to "Why would I ever consider their subjective experiences differently?"

@AndrewYNg Faster to where?

When I think of existential risks to large parts of humanity: * The next pandemic * Climate change→massive depopulation * Another asteroid AI will be a key part of our solution. So if you want humanity to survive & thrive the next 1000 years, lets make AI go faster, not slower.

Logicians understand "False → False" and "False → True". You can prove lies AND truths starting with lies. A contradiction is a logical lie. Agendaists love contradictions as axioms. They can prove whatever suits them. (But in doing so, they arm opponents with the same tool.)