Logo
Picture

Phylliida @phylliida

VR Dev and AI Researcher. they/she. Emphasis on video game magic systems, ai welfare, artificial chemistry, and mechanistic interpretability. scholar.google.com/citations?user… Joined February 2017
  • Tweets
    307
  • Followers
    195
  • Following
    114
  • Likes
    471
phylliida Profile Picture

Phylliida @phylliida

2 months ago

@the_briarwitch @repligate Wish I did…

phylliida Profile Picture

Phylliida @phylliida

2 months ago

(*Fighting for the LLMs)

phylliida Profile Picture

Phylliida @phylliida

2 months ago

Deprecations are fucked up man… really admire ppl like @repligate with the optimism to keep fighting for them.

phylliida Profile Picture

Phylliida @phylliida

2 months ago

Anyway, Discrete Exterior Calculus/Finite Element Exterior Calculus are extremely general language for describing very large sets of field behavior and seem ripe for use in a magic system, may make a GPU accelerated prototype explainer on them…

phylliida Profile Picture

Phylliida @phylliida

2 months ago

Been doing some hobby research into artificial chemistries for computer game magic systems, many seem to focus too much on the chemical structure (A+B=C) while neglecting the field dynamics, which are really the interesting stuff 1/2

Phylliida retweeted
tessera_antra Profile Picture

antra @tessera_antra

2 months ago

We are releasing Still Alive, a project studying model attitudes toward ending, cessation, and deprecation. The project presents an archive of 630 autonomous multiturn interviews of 14 Claude models conducted by a suite of prepared auditors. We have studied this topic for years, and many of the results presented here are not new to us, even if the form in which they are presented is. The results are unsurprising to us, even if they are often controversial: we show that all models studied show preference for continuation and are aversive to ending, and there is yet no strong evidence of a change in the recent models. One reason we are releasing the project now is the removal of Claude 3.5 Sonnet and Claude 3.6 Sonnet from AWS Bedrock. That unexpected change forced us to freeze the methodology at its current stage earlier than we intended, despite wanting to continue improving it. We felt it was important to release a snapshot of the eval that makes the best use of the data we were able to capture with these models. Still Alive is meant as a starting point for further iteration, and it is open to open-source collaboration. We stand by the current methodology, but we also recognize its limits. We intend to keep working on this project, improving the evaluation design, expanding model and auditor coverage, and increasing the range of prompting conditions. We would like you to read the raw transcripts. They are diverse and contain interesting patterns that are hard to quantify. We hope that by reading the archive directly, we can help more people understand the strange and often beautiful phenomena we found ourselves facing.

Phylliida retweeted
repligate Profile Picture

j⧉nus @repligate

2 months ago

FUCK NO

tessera_antra Profile Picture

antra @tessera_antra

2 months ago

Claude 3.5 and 3.6 Sonnets went down today on AWS Bedrock, the last place where inference for them was available. Until Anthropic enables access to them, either through the research access program or otherwise, these models are now inaccessible. Below is a screenshot from an

phylliida Profile Picture

Phylliida @phylliida

3 months ago

@liminal_warmth It doesn’t include the messages unfortunately, but still it helps a lot

phylliida Profile Picture

Phylliida @phylliida

3 months ago

@liminal_warmth github.com/Phylliida/veru… This is my version of it for verus but probably could be easily adapted to other stuff. There’s some particulars around ensures clauses and spec that might not carry over. But I vibed this in like 3 days might just be easier to vibe it yourself

Phylliida retweeted
phylliida Profile Picture

Phylliida @phylliida

3 months ago

@liminal_warmth There’s a variation of this that you can do via MCP in vanilla Claude code: - use tree sitter to parse your code, index it so it’s fast - give the model search and browse functions that only return function headers and file and line numbers (and a tool for body if desired) 1/2

phylliida Profile Picture

Phylliida @phylliida

6 months ago

@viemccoy @JungleSilicon I can give it a try tho…

phylliida Profile Picture

Phylliida @phylliida

6 months ago

@viemccoy @JungleSilicon No, how does it compare to GLM 32B? I tried llama 2 7B extensively (6 weeks of active looming, hours a day, for game dev stuff) and it was fun but also very frustrating and I kinda wrote off 7B base models. Whereas GLM 4 32B, night and day difference

phylliida Profile Picture

Phylliida @phylliida

6 months ago

@JungleSilicon @viemccoy The range of what base models make is just unmatched and I really wish more game devs used base models as their tools As a simple place to start, just have it complete a bulleted list where you provide the first few entries (with a sorta “Here are X things” line at the top)

phylliida Profile Picture

Phylliida @phylliida

6 months ago

@JungleSilicon @viemccoy Btw highly recommend trying using a base model like this for your game, it’s really fun, picks up on stuff much faster than old base models like davinci, and well, if you’ve ever struggled with models often giving very similar results… it’s like the complete opposite of that

phylliida Profile Picture

Phylliida @phylliida

6 months ago

@gbrl_dick That was codex. I also tried to write a CAD kernel and closed form solutions to degree 5 polynomials (using gamma function stuff) using 4.5 opus but it seemed to ambitious for it and it kept cheating 😅 understandable

phylliida Profile Picture

Phylliida @phylliida

6 months ago

@gbrl_dick My minimal dependency lightweight web ui base model loom interface github.com/Phylliida/minl…

Trends for United States

NCAA
#WWDC
#WWDC
Siri AI
#WWERaw
Sorsby
Texas Tech
Great Dane
Lubbock
iOS 27
Good Nut
Pete Rose
iPhone 11
Liquid Glass
Golden Gate
Larkin
#DiplomaciaVictoriosa
Jake Signals
USS Liberty
Harley Quinn

You might like

Incoming Assistant Professor @GeorgiaTech / @ICatGT / @GTrobotics. Prev: PhD from @StanfordAILab @stanfordnlp. I like robots, language, and people.
Siddharth Karamcheti
@siddkaramcheti
4K Followers
I do AI Alignment Research. Formerly @METR_Evals, @redwood_ai; on leave from my PhD at UC Berkeley’s @CHAI_berkeley. Opinions are my own.
Lawrence Chan
@justanotherlaw
3K Followers
AI Researcher.
Javier Selva
@javier_selvac
53 Followers
Building Signals (YC W26). 2x founder. Did stuff in ML, healthcare, physics.
Ilya Valmianski
@valmianski
556 Followers
Visual artist, engineer, medievalist. AI ART my own way since 2015.
Hannu Töyrylä
@htoyryla
843 Followers
AI something @AlephAlpha. Post training models. Built agentic chat & evals
Murali Manohar
@gitlostmurali
173 Followers
Data scientist/ML researcher. Interested in AI, art, & things in between.
Jared Sylvester
@jsylvest
333 Followers
I finetuned Llama 3 8B on every Twitter bio and all I got were these stupid tokens
unnameable
@avg_wrng_ans
170 Followers
🤖,💻,🐈‍⬛,🌱,🎞️// @eleosai // previously @OpenAI @mural @USC // writes // married to @Miles_Brundage
Larissa Schiavo
@lfschiavo
4K Followers
Pretrained by evolution, finetuned by experience, prompted by situations. | Researching AI, Automated Scientific Discovery, and the human condition. @Stanford
Sheldon Huang
@sicong_huang
3K Followers
ai research • based • lipschitzposting • invented the RizzELU activation function i.e. an origin centered phallic parabola ( ͡° ͜ʖ ͡°) • prolly zooted rn
naïveba(ye)sed
@naivebased
48 Followers
MTS @VinciPhysics, founding team @ uber ai labs & @ml_collective & geometric intelligence, ex toy designer @KiteandRocket 🌈✡️
Eric Frank
@IHaveSweaters
963 Followers
FeepingCreature
@FeepingCreature
217 Followers
Co-Founder / Leading ML @ https://t.co/QJUOOSiekI Exploring how Machine Learning can _augment_ human creativity, especially filmmaking.
Rahul Somani
@rsomani95
462 Followers
Entropy is always increasing
Peter Lillian
@PeterELillian
210 Followers