|
01:43
japhb left
01:49
japhb joined
|
|||
| Voldenet | LLMs are good only if you run it for free (essentially only making llm companies lose money) but don't let people know that you've used it }:-> | 02:21 | |
| that's the approach I recommend and use | |||
| Ethically all llms are putrid and taught by tons of stolen (or lawfully stolen – authors didn't know they agreed to that) media | 02:24 | ||
| so the only sane approach is to never pay for their services and see them bleed invested money | 02:27 | ||
| disbot4 | <aruniecrisps> @timo @Voldenet I work with LLMs on the regular, and I'm forced to work with them. Not a fan of them at all | 04:39 | |
| Voldenet | I understand because I was forced to implement support for LLMs at work, these are sad times | 05:21 | |
| but I am not forced to write code using LLMs fortunately | |||
|
07:48
rnddim joined,
ShimmerFairy is now known as Guest4078,
Guest4078 left,
rnddim is now known as ShimmerFairy,
ShimmerFairy left
07:49
ShimmerFairy joined
07:54
ShimmerFairy left,
ShimmerFairy joined
|
|||
| disbot4 | <librasteve> I only started serious playing with LLMs in the last month or so - I have had modest success using Claude with Raku - it probably makes me 50-150% more productive. I only use it where I choose to and I would hate it if forced to use in a work setting. I don't like the morality (copyright theft, pathetic rationale for layoffs you were going to do anyway). can see projects becoming viable in reasonable time for a single coder | 12:37 | |
| or a small team - Apache Arrow for Raku? | |||
| <librasteve> Anyway - thanks to all those who have shared your views and I understand that this evokes some very strong feelings and opnions. | 12:38 | ||
| lizmat | .oO( fried opnions, hmmmm! :-) |
||
| disbot4 | <librasteve> I am inserting deliberate erors to assert my humanness | 12:39 | |
| lizmat | sorry, in a weird mood because of the passing of one of my hearoes: en.wikipedia.org/wiki/Wim_T._Schippers :-) | 12:40 | |
| Pollens! | |||
| disbot4 | <aruniecrisps> @librasteve my overall opinion on LLMs is basically the same as Andrew Kelley of Zig, i think he lays it out perfectly | 13:09 | |
| <librasteve> yeah, I think I read something about that - maybe that's a better model the the LLVM approach I started with in github.com/Raku/problem-solving/issues/510 ... any chance you can find a link and add it to that thread (although tbh I feel that trying to set a policy may be premature and best that the small team just finds a practical way to work). | 13:25 | ||
|
17:24
japhb left
|
|||
| Geth | MoarVM/larger_in_situ_strings: 4fd6475b9c | (Timo Paulssen)++ | 8 files MVMString: Move num_strands into storage union it's only needed for strand storage type in the first place, and putting it into the union means the otherwise wasted two bytes of space after the other pointer storage types blob_32 and blob_8 can actually be used to store two more 8bit graphemes in in situ 8bit storage mode. This commit also changes all occurences of the magic value 8 to instead use a define that can be used at compile time to make the in situ storage for 8bit graphemes even bigger, although the in situ 32bit storage doesn't increase if you make it large enough to have more there. |
18:10 | |
| MoarVM/larger_in_situ_strings: f586798090 | (Timo Paulssen)++ | tools/libmoar.so-gdb.py follow MVMString structure change in gdb plugin |
18:18 | ||
| MoarVM/build_fuzzing_targets: 24 commits pushed by (Timo Paulssen)++ review: github.com/MoarVM/MoarVM/compare/f...c0c472a122 |
18:20 | ||
| timo | (^- more than just a rebase!) | 18:21 | |
| Geth | MoarVM/build_fuzzing_targets: 9360b0ae9f | (Timo Paulssen)++ | 8 files Skip CUs or named files from bb edge coverage with env vars MVM_COVERAGE_SKIP_CU_FILTER and MVM_COVERAGE_SKIP_FILE_FILTER allow setting multiple part-of-names to skip whole CUs or individual frames based on the filename. Next step should be a allowlist in addition (or instead of) the denylist. |
18:53 | |
| disbot4 | <aruniecrisps> @librasteve Here is Andrew Kelley's take on the matter: www.businessinsider.com/zig-progra...les-2026-5 | 19:08 | |