@archo@mastodon.gamedev.place

Arvīds Kokins @archo@mastodon.gamedev.place

· 7mo ago

normalize making custom licenses, they make corporate lawyers nervous

there's seemingly a funny side effect for this one btw:

> 2. No part of a work under this license may be used or reproduced in any manner by organizations or businesses with more than fifty (50) employees.
> The Filthy Human Hands License is licensed under the Filthy Human Hands License.

so any org with >50 employees can't even copy the license without immediately needing fair use exceptions or equivalent

#licensing

infosec.exchange

Cat 🐈🥗 (D.Burch): "The Filthy Human Hands (FHH) License v1.0 https:…" - Infosec Exchange

107

1

65

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

https://gamedev.net/portfolios/2560-neon-shaodaowera this shit is in the portfolio, categorized as a "game" what exactly is being showcased here? prompting? some of the images even have random generated watermarks on them under the gemini watermark lol #gamedev #ai #slop

gamedev.net

1

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

@frankreiff @ikuturso @chaucerburnt @sinbad on explaining past actions however, I see humans as having some distinct advantages:

operating with an extremely broad range of conceptual units (unlike LLM wordpiece tokens) for highly efficient encoding
we can reprioritize notability of observations based on past events such that if we end up regretting missing some fact, we can figure out that we need to pay more attention to something in the future, which doesn't seem possible with LLMs

0

2

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

@frankreiff @ikuturso @chaucerburnt @sinbad

we have the ability to operate with order-sensitive concepts like time and operational preconditions such that we can fill in the gaps from our extremely flawed memories with reasoning about what is possible and what isn't, as well as combining different relative orders of events into a single timeline

LLMs seem completely incapable of any of that, and even the ordering of input tokens appears to be a mere suggestion and thus unreliable

0

1

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

visual studio has this almost decent thing where only the most recent tabs are visible but the rest are pushed into a list

however

the order management of the visible tabs is random unpredictable shit when new tabs are opened
list ordering is alphabetical, not MRU
the list is not searchable
the list does not show which items are already shown in the tabs list
the list only shows the opened items, for anything else you need to go to an entirely different list

#ui

0

1

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

they used to easily pull 100k+ views even a couple of years ago: https://gamedev.net/blogs/?q=&sort=popular the "trending" posts from the last month or so get less than 2000, or basically what a complete noname creator could pull from reddit/tiktok by posting something random a few times #gamedev

gamedev.net

0

2

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

the reason why I'm looking into this less efficient form of interpolation is that I want to combine it with angular velocity for my quadratic spline thing the problem with quaternions being that interpolating quaternions does not map to any kind of physical behavior at the numerical level whereas axis-angle is angular velocity vec3 dir=axis, length=angle encoding also allows componentwise separation such that each component can get a separate curve #gamedev #math

0

4

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

I clicked through a couple of the bugs they posted and all of the ones I saw at least were single-function bugs (some involved recursion) with some extremely basic mistake patterns e.g. failing to make a dependent change conditional on whether a function call succeeds, or comparing the wrong pointer out of two idk what their analysis infra is like but based on the output, I guess the LLM found the simple patterns first and then walked the code backwards to find the triggering condition

0

2

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

what if it was possible to instead reorder the list of openable things by MRU (most recently used) by default, show opened file indicators there as well, allow closing only via the context menu, and ditch tabs entirely? probably wouldn't work for things like browsers necessarily, since there are tons of per-tab context that can't all be serialized (e.g. open websocket connections) - but might work better for editors #ui

0

1

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

the projects page is full of slop images: https://gamedev.net/projects/ either that or programmer art (but both are kind of the same thing) there's nobody (individuals or teams) with enough skill to make anything coherent on their own on that website #gamedev #slop

gamedev.net

0

2

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

walking the code backwards however is something static analysis tools have been able to do for some time so it's not even necessary for an LLM to be involved for that part (at least for static languages) so I would expect that eventually (if not already) that part will be made deterministic, the initial pattern matching could use a smaller and specialized "model" operating with normalized code (removing syntax and spelling from the equation) and only cross-lang exploit gen may still need an LLM

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

given real starting/ending rotations once the temporary starting rotation is expressed as [0;0;0] and the temporary ending rotation is the difference of real starting/ending rotations, expressed as vec3 dir=axis, length=angle it is now possible to do a componentwise interpolation between the rotations using the QLIF method (https://mastodon.gamedev.place/@archo/116516977975559261) because it requires finding an intersection point between 2 straight lines creating an editor for that will be a challenge though #gamedev #math

mastodon.gamedev.place

Arvīds Kokins: "here's an implementation: https://github.com/arch…" - Gamedev Mastodon

0

3

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

one of the editing challenges is that full-circle loops will -not- start and end in the same location, if curves need to be shown the curves require picking any key"frame" as a reference point which will be [0;0;0], and then build [vec3 dir=axis,length=angle] points in both directions relative to their directionally previous points one full turn would be 360 degrees away vertically this is consistent with all angle-based interpolations but not quaternions #gamedev #math

0

2

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

here's a small prototype of the idea described in https://mastodon.gamedev.place/@archo/116537863609460739 not entirely sure if it works in all cases yet but it seems to work in a simple test case I tried to show interpolation evenly by hand so you can see the effect start/end velocity has on the interpolation #gamedev #math

mastodon.gamedev.place

Arvīds Kokins: "given real starting/ending rotations once the te…" - Gamedev Mastodon

0

2

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

one of the unintuitive things that can be observed is that the Z axis has a start/end angle difference even though seemingly none of it's been defined that's because UI angles are Euler angles in XYZ order but the second curve point is the difference between start/end rotations in dir=axis,length=angle format in this case, going from euler xyz [80;0;0] to euler xyz [0;80;0] creates a difference of roughly [-66;66;55] #gamedev #math

0

1

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

@tojiro arguably they did solve "reviewing other people's code" too :)

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

@frankreiff @ikuturso @chaucerburnt @sinbad LLMs definitely don't imitate any natural structures at any level, as AFAIK there's no equivalent to floating point values, tokens or positional encoding etc. and I'm not sure it is necessarily sufficient for intelligence to only have a brain - I'm not aware of any creature that has become smart by existing as a disembodied brain so there isn't really much reason to assume the contribution of all the surrounding structures isn't relevant enough

0

Arvīds Kokins @archo@mastodon.gamedev.place

· 5mo ago

Replying to

@sibaku@mas.to yeah, from a conceptual perspective it's quite clear that they're supposed to be doing the same thing but from a mathematical one (as in, the actual sequence of operations) they seem to be doing very different things 🤷 that's what is a bit confusing to me but given the conceptual equality, I expect there to be a series of steps (mathematically equivalent reformulations) that go gradually from one to the other

0