Arvīds Kokins
@archo5.bsky.social
Trying to make things that vaguely resemble games (and the tech to make them work). Recently released corpo/ghost as @interleavedsystems.com
https://store.steampowered.com/app/2844660/corpoghost/?utm_source=bskybio
https://store.steampowered.com/app/2844660/corpoghost/?utm_source=bskybio
also new texture compression tech was released last year, so I wonder if zuck's announcement is in any way related to that (like, if they got some internal demo to work or something like that)
www.ludicon.com/castano/blog...
www.ludicon.com/castano/blog...
Spark 1.3 and spark.js – Ignacio Castaño
www.ludicon.com
January 30, 2026 at 7:23 PM
also new texture compression tech was released last year, so I wonder if zuck's announcement is in any way related to that (like, if they got some internal demo to work or something like that)
www.ludicon.com/castano/blog...
www.ludicon.com/castano/blog...
I hope that's realtime 3D specifically he's dumping money into
as someone working with realtime 3D, I approve
I mean, it still sounds quite stupid but I'd take his money or the money of anyone trying to mindlessly copy him lol
as someone working with realtime 3D, I approve
I mean, it still sounds quite stupid but I'd take his money or the money of anyone trying to mindlessly copy him lol
January 30, 2026 at 7:15 PM
I hope that's realtime 3D specifically he's dumping money into
as someone working with realtime 3D, I approve
I mean, it still sounds quite stupid but I'd take his money or the money of anyone trying to mindlessly copy him lol
as someone working with realtime 3D, I approve
I mean, it still sounds quite stupid but I'd take his money or the money of anyone trying to mindlessly copy him lol
p.s. I've explained why that shit is doomed the last time crap like that made waves:
bsky.app/profile/arch... 🧵
the great thing with fundamentals is that no matter how many fancy scam demos someone ships, the fundamentals won't ever change
bsky.app/profile/arch... 🧵
the great thing with fundamentals is that no matter how many fancy scam demos someone ships, the fundamentals won't ever change
They could increase the resolution or collect more data or track game state.
They can't make this without making the actual game first.
They can't (statistically) compress untold amounts of data into a much smaller fixed amount of space without losses.
They can't escape combinatorial explosions.
They can't make this without making the actual game first.
They can't (statistically) compress untold amounts of data into a much smaller fixed amount of space without losses.
They can't escape combinatorial explosions.
January 30, 2026 at 7:00 PM
p.s. I've explained why that shit is doomed the last time crap like that made waves:
bsky.app/profile/arch... 🧵
the great thing with fundamentals is that no matter how many fancy scam demos someone ships, the fundamentals won't ever change
bsky.app/profile/arch... 🧵
the great thing with fundamentals is that no matter how many fancy scam demos someone ships, the fundamentals won't ever change
it's kind of the other way around - ML needs lots of explicit examples, rules can be as vague or detailed, or as broad or specific as you need
in my view, the real benefit of ML relative to a rules-based system is the massively reduced cost per rule/example, since rules need expensive coder time
in my view, the real benefit of ML relative to a rules-based system is the massively reduced cost per rule/example, since rules need expensive coder time
January 30, 2026 at 12:04 PM
it's kind of the other way around - ML needs lots of explicit examples, rules can be as vague or detailed, or as broad or specific as you need
in my view, the real benefit of ML relative to a rules-based system is the massively reduced cost per rule/example, since rules need expensive coder time
in my view, the real benefit of ML relative to a rules-based system is the massively reduced cost per rule/example, since rules need expensive coder time
yeah, though I wasn't talking just about typos
an LLM has to predict all the inputs that might appear, otherwise they may well detect "turn off the TV" as "turn off the light", if they only knew about the latter
an LLM has to predict all the inputs that might appear, otherwise they may well detect "turn off the TV" as "turn off the light", if they only knew about the latter
January 30, 2026 at 11:59 AM
yeah, though I wasn't talking just about typos
an LLM has to predict all the inputs that might appear, otherwise they may well detect "turn off the TV" as "turn off the light", if they only knew about the latter
an LLM has to predict all the inputs that might appear, otherwise they may well detect "turn off the TV" as "turn off the light", if they only knew about the latter
anything falls over when it encounters an unpredicted input, that just seems to be a fundamental feature of all AI developed to date (or even all software), not specific to a rules-based system
glitch tokens make LLMs fall over
glitch tokens make LLMs fall over
January 30, 2026 at 11:50 AM
anything falls over when it encounters an unpredicted input, that just seems to be a fundamental feature of all AI developed to date (or even all software), not specific to a rules-based system
glitch tokens make LLMs fall over
glitch tokens make LLMs fall over
I didn't say they copy and paste their input, and did say that they detect correlated sequences, so I'm not sure where I could have oversimplified
they do have a best guess but the flipside of it is that you also need many negative examples to prevent LLMs from guessing extremely incorrectly
they do have a best guess but the flipside of it is that you also need many negative examples to prevent LLMs from guessing extremely incorrectly
January 30, 2026 at 11:39 AM
I didn't say they copy and paste their input, and did say that they detect correlated sequences, so I'm not sure where I could have oversimplified
they do have a best guess but the flipside of it is that you also need many negative examples to prevent LLMs from guessing extremely incorrectly
they do have a best guess but the flipside of it is that you also need many negative examples to prevent LLMs from guessing extremely incorrectly
the main problem with rules-based models is managing the complexity (handling typos/variations and tweaking the probabilities of various imperfect matches)
there is no theoretical limit to how well they could parse, whereas LLMs are strictly limited by the totality of examples compressed within
there is no theoretical limit to how well they could parse, whereas LLMs are strictly limited by the totality of examples compressed within
January 30, 2026 at 11:33 AM
the main problem with rules-based models is managing the complexity (handling typos/variations and tweaking the probabilities of various imperfect matches)
there is no theoretical limit to how well they could parse, whereas LLMs are strictly limited by the totality of examples compressed within
there is no theoretical limit to how well they could parse, whereas LLMs are strictly limited by the totality of examples compressed within
saying it "parses" or "follows rules" seems like a stretch as well
if a specific use doesn't exist in the data - in whole or in parts, including common typos, and with enough variations to distinguish exact sequences from correlated sequences - it won't be recognized correctly
if a specific use doesn't exist in the data - in whole or in parts, including common typos, and with enough variations to distinguish exact sequences from correlated sequences - it won't be recognized correctly
January 30, 2026 at 11:26 AM
saying it "parses" or "follows rules" seems like a stretch as well
if a specific use doesn't exist in the data - in whole or in parts, including common typos, and with enough variations to distinguish exact sequences from correlated sequences - it won't be recognized correctly
if a specific use doesn't exist in the data - in whole or in parts, including common typos, and with enough variations to distinguish exact sequences from correlated sequences - it won't be recognized correctly
the only way I imagine a TTS could be truly ethical is if the data is solely owned by the owner of the voice, and is only used while the real voice is unavailable and its owner is still alive (to avoid both unauthorized use and shortchanging end users of a vocal performance)
not impossible but hard
not impossible but hard
January 30, 2026 at 11:20 AM
the only way I imagine a TTS could be truly ethical is if the data is solely owned by the owner of the voice, and is only used while the real voice is unavailable and its owner is still alive (to avoid both unauthorized use and shortchanging end users of a vocal performance)
not impossible but hard
not impossible but hard
the name isn't very accurate, "LLMs" are lossy solid number sequence compressors
it doesn't "understand language" and doesn't even have much to do with language really
as for a small set of simple text commands, code that recognizes those could be done by hand, idk if LLMs add much value there
it doesn't "understand language" and doesn't even have much to do with language really
as for a small set of simple text commands, code that recognizes those could be done by hand, idk if LLMs add much value there
January 30, 2026 at 11:14 AM
the name isn't very accurate, "LLMs" are lossy solid number sequence compressors
it doesn't "understand language" and doesn't even have much to do with language really
as for a small set of simple text commands, code that recognizes those could be done by hand, idk if LLMs add much value there
it doesn't "understand language" and doesn't even have much to do with language really
as for a small set of simple text commands, code that recognizes those could be done by hand, idk if LLMs add much value there
at the end of the day, everything becomes oversaturated, at which point it's time to move on, except there's increasingly less new things to move on to, and we circle back to the old
I keep wondering how many flavored repeats humanity still has in it until we start self-sabotaging out of boredom
I keep wondering how many flavored repeats humanity still has in it until we start self-sabotaging out of boredom
January 29, 2026 at 8:20 PM
at the end of the day, everything becomes oversaturated, at which point it's time to move on, except there's increasingly less new things to move on to, and we circle back to the old
I keep wondering how many flavored repeats humanity still has in it until we start self-sabotaging out of boredom
I keep wondering how many flavored repeats humanity still has in it until we start self-sabotaging out of boredom
highlights by definition escape the average of human activity
e.g. as more people get to travel for fun, the topic becomes oversaturated and uninteresting
it used to be a life-changing adventure with many risks, now it's often an industrial style pipeline with all the rough edges sanded off
e.g. as more people get to travel for fun, the topic becomes oversaturated and uninteresting
it used to be a life-changing adventure with many risks, now it's often an industrial style pipeline with all the rough edges sanded off
January 29, 2026 at 8:09 PM
highlights by definition escape the average of human activity
e.g. as more people get to travel for fun, the topic becomes oversaturated and uninteresting
it used to be a life-changing adventure with many risks, now it's often an industrial style pipeline with all the rough edges sanded off
e.g. as more people get to travel for fun, the topic becomes oversaturated and uninteresting
it used to be a life-changing adventure with many risks, now it's often an industrial style pipeline with all the rough edges sanded off
as a side note, it's a bit depressing to think that the highlights of most people's lives probably wouldn't be enough to carry a 2h podcast
not that it's important in any way, just the deep misalignment of actual value created vs what seems superficially valuable to others as entertainment
not that it's important in any way, just the deep misalignment of actual value created vs what seems superficially valuable to others as entertainment
January 29, 2026 at 8:02 PM
as a side note, it's a bit depressing to think that the highlights of most people's lives probably wouldn't be enough to carry a 2h podcast
not that it's important in any way, just the deep misalignment of actual value created vs what seems superficially valuable to others as entertainment
not that it's important in any way, just the deep misalignment of actual value created vs what seems superficially valuable to others as entertainment
at the same time, I've been seeing podcasts going increasingly deeper into arguably more niche subjects with some fairly introspective/knowledgeable guests
most of it is probably not going to break out into anything larger but something easily could (hard to predict what actually will tho)
most of it is probably not going to break out into anything larger but something easily could (hard to predict what actually will tho)
January 29, 2026 at 7:58 PM
at the same time, I've been seeing podcasts going increasingly deeper into arguably more niche subjects with some fairly introspective/knowledgeable guests
most of it is probably not going to break out into anything larger but something easily could (hard to predict what actually will tho)
most of it is probably not going to break out into anything larger but something easily could (hard to predict what actually will tho)
a side effect for the guests seems to be that in the rush to capitalize on their personal brand, they may have revealed a bit too much about their true abilities
a lot of them just don't seem to have that much going on between the ears, or in their lives
a lot of them just don't seem to have that much going on between the ears, or in their lives
January 29, 2026 at 7:53 PM
a side effect for the guests seems to be that in the rush to capitalize on their personal brand, they may have revealed a bit too much about their true abilities
a lot of them just don't seem to have that much going on between the ears, or in their lives
a lot of them just don't seem to have that much going on between the ears, or in their lives
wondering if it will rebranded/restructured into something again (like what podcasts were in comparison to radio) or if it'll just stay mostly flat
this time around it seems like they've already been recut and edited for every mode of consumption imaginable, including several levels of reactions
this time around it seems like they've already been recut and edited for every mode of consumption imaginable, including several levels of reactions
January 29, 2026 at 7:47 PM
wondering if it will rebranded/restructured into something again (like what podcasts were in comparison to radio) or if it'll just stay mostly flat
this time around it seems like they've already been recut and edited for every mode of consumption imaginable, including several levels of reactions
this time around it seems like they've already been recut and edited for every mode of consumption imaginable, including several levels of reactions
Humans& 🤨
Humansand 🤔
Human...sand 😱
vibe:
Humansand 🤔
Human...sand 😱
vibe:
January 28, 2026 at 3:23 PM
Humans& 🤨
Humansand 🤔
Human...sand 😱
vibe:
Humansand 🤔
Human...sand 😱
vibe:
Beidzās otrais saskaņošanas termiņš:
tapportals.mk.gov.lv/legal_acts/4...
Dažviet var redzēt, ka atšķiras viedoklis starp pirmo un otro termiņu (piem. FLA un mogo).
tapportals.mk.gov.lv/legal_acts/4...
Dažviet var redzēt, ka atšķiras viedoklis starp pirmo un otro termiņu (piem. FLA un mogo).
Par vienotas patērētāju kreditētāju licencēšanas un uzraudzības institūcijas noteikšanu
tapportals.mk.gov.lv
January 27, 2026 at 9:16 AM
Beidzās otrais saskaņošanas termiņš:
tapportals.mk.gov.lv/legal_acts/4...
Dažviet var redzēt, ka atšķiras viedoklis starp pirmo un otro termiņu (piem. FLA un mogo).
tapportals.mk.gov.lv/legal_acts/4...
Dažviet var redzēt, ka atšķiras viedoklis starp pirmo un otro termiņu (piem. FLA un mogo).
if we're just saying numbers, why not $69?
January 26, 2026 at 7:14 PM
if we're just saying numbers, why not $69?
"feel"? "grow up"?
what are the physical processes that facilitate any of that?
you'd think that in a company full of (arguably) people who are supposed to understand the tech they're making, she could've found a single person to explain to her how it actually works
what are the physical processes that facilitate any of that?
you'd think that in a company full of (arguably) people who are supposed to understand the tech they're making, she could've found a single person to explain to her how it actually works
January 26, 2026 at 5:40 AM
"feel"? "grow up"?
what are the physical processes that facilitate any of that?
you'd think that in a company full of (arguably) people who are supposed to understand the tech they're making, she could've found a single person to explain to her how it actually works
what are the physical processes that facilitate any of that?
you'd think that in a company full of (arguably) people who are supposed to understand the tech they're making, she could've found a single person to explain to her how it actually works