Loading activity…
@gowthami_s
Model tasting @theworldlabs | SGD-ing on life | Rants, Memes, Opinions my own.
AI
A user says they passed on tweeting about two papers they viewed as “pretty much p-hacked” rather than scalable results.
AI
A commenter says early-only training scored better on motion than late-only training, which did better on anatomy.
AI
A user says 2B text-to-video DDRL trails DanceGRPO on VideoAlign reward but leads human preference by 10.5 percentage points.
AI
A user says they need an LLM to decode their credit card transactions and asks who is building one.
AI
A post says RVM defines its target from the final sample rather than noisy intermediate transitions.
AI
A user says looped transformers are all over their feed, recalling a 2016 paper where they first saw neural-network looping.
AI
A reply credits @RulinShao for two papers on the list; the credited author celebrates the result.
AI
A researcher describes a spatially localized 3D reward model designed to show where generated scenes go wrong.
AI
A commenter says the paper leaves “single-shot accuracy” unclear and omits key evaluation settings.
AI
A user predicts that innovations in synthetic data will bring the next step change in technology.
AI
Posts on X congratulated Nathan Lambert and Trillium Labs after a launch, with several users describing the effort as open, science-oriented or focused on open post-training. Lambert later said some criticism came from Substack and more general-audience AI followers, not the technical AI community.
AI
A builder says letting friends try work-in-progress models this way has brought quicker feedback and faster iteration.
AI
A user argues that Meta-FAIR’s Galactica faced a backlash that pushed its team to remove the demo, contrasting that with the enthusiastic welcome ChatGPT later received.
AI
A post frames the relationship between people and large language models as a question of control.
AI
A researcher urges junior researchers to hone their writing and think carefully about what individual words mean and evoke in readers.
AI
ICLR says its 2027 conference will feature two days of workshops in San Francisco and invites researchers to submit proposals to organize them.
AI
An opencode2 developer says its codemode can display a user interface and prompt for permissions even when a model sends over a chunk of code.
AI
The World Labs researcher posted criticism of Astra alongside a meme image.
AI
Gowthami Somepalli shares her prompt asking Astra to reinvent her without errors.
AI
World Labs multimodal AI researcher Gowthami Somepalli posted about her initial assignment.
