env impact is same ballpark as streaming video, even when training cost is amortized over end users.
so: i understand the env argument. but it's wrong/hypocritical.
RichTextLabel's Fit Content flag either doesn't work, or i don't understand how it does work. either way: this isn't an expected result for me (min size isn't being auto adjusted to make sense w content)
other use cases:
you're NASA and you need to plan navigation routes for a Mars rover.
you want to solve open ErdΕs problems.
you want to make a breakthrough in theoretical physics.
llms are currently doing these things.
It's hard to convey how nuts it feels when just about every few weeks to a month tops, some perfectly acceptable research establishes LLMs can do mental reasoning/metareasoning tasks critics say are impossible, and not only does it fail to register - it leads to increasingly aggressive dismissals
works just the same way as the regular matrix, with a node selected you can toggle any of the cells to assign occupancy/scanning
yes, the hardware isn't free. I agree.
interesting framing!
relatively less valuable with respect to experienced devs, yes. because llms are power tools/capability multipliers.
lack of wealth comes w serious disadvantages, to be sure. this doesn't equal neglect on the part of employers.
you havent understood me (and mischaracterised my position). wrt employer decisions expected productivity is paramount. you don't employ scribes when a printing press is available for the same reason that llm proficient coders might be preferred.
but that can be true: insufficient wealth can mean that you must use old fashioned means.
'human first' work would also council employing scribes rather than using the printing press. this approach would be disastrous for human welbeing.
you're not specifying high enough min iterations on yr ralph loops smh.
I've had the same thing. then the LLM is like my bad you're not dying, it's actually within the normal distribution.
π€·
as a book smart kid, no, I don't think so.
fair.
but what's the implication of the 'neglectful' part? have people *not* use or develop this incredible cost/time saving tech because it's not free?
9 months was the latest estimated lag time between closed SOTA model capability and that of open models, that can be run 'for free'. so this isn't leaving behind it's trailblazing, the free versions then follow.
the people yearn for different flavours of pause
making a scoped pause system for godot projects: scopes define collections of nodes to pause. different systems/events (e.g. dialogue, cutscene, hitstop, level-win) can pause the same scope using different activation_ids. a scope only unpauses when the last activation releases.
ok, but that profile doesn't match the person mentioned in this thread
my code is *much* more maintainable, well documented, and robust since I started using llms. it's day and night. llms have infinite patience for housekeeping, hardening, and streamlining.
principled actors can coordinate acausally, while actors that merely align based on loyalty must negotiate
"LLM code is low quality" is a self-confession in the first place
like I have this thing hammering through mutants tests right now and finding bugs I one thousand percent never would have found or bothered to fix by hand, because I do not have infinite time and attention span for it
the highlighted section is absolutely false.
this is wrong. llms don't output the average of [some category of text].
a prompt/context is a coordinate. stand in a crowded part of the map (generic prompt): get generic continuations. stand in a narrow pass w hard constraints (unusual & specific context) and u end up somewhere more interesting.
awkward thing: for my work codex is faster than opus, more skilled, and unlike opus I've never hit token limits. for ~ the same subscription cost.
I agree that 'agent' isn't a great term to denote llms that can read/write files and run scripts (vs those that can't). it suggests more independence. and we'll need a term for that soon.
also might be worth trying: a skill that uses RAG over a corpus of target writing, dynamically surfaces semantically similar sections as reference for the new passage
someone mischaracterising the nature of the disagreement, claiming youd made a claim to the effect of they 'think like humans', and that you failed in the demonstration.
Recently I got invited to join the new Godot UX team. When I have time I will be helping testing editor PRs and make Godot more usable for everyone.
What are some of the most annoying things about Godot UI/UX that you think need attention?
