Sabitlenmiş Tweet
fun fact: we're actually a research company
Paradigma@paradigmainc
your research doesn’t have to be ugly. today we’re releasing a big update to the flywheel graph canvas. try it out on flywheel dot paradigma dot inc
English
tensorqt
8.1K posts
tensorqt
@tensorqt
⬢ chaos dancing star @paradigmainc
Icecrown Citadel Katılım Şubat 2022
401 Takip Edilen5.6K Takipçiler
@kirbytheodor @zhengyaojiang @paradigmainc this is very good work (we have had some very similar results). We will support an abstraction to represent this idea naturally in Flywheel soon
English
@zhengyaojiang have you considered integrating and or comparing this with flywheel by @paradigmainc ? I see a lot of parallels with the looping ideology and graph structure
English
The first experimental evidence of recursive self-improvement (RSI).
Autoresearching the autoresearch agent for eight days.
The result beats the harness we hand-tuned for two years, on held-out benchmarks: 🧵(1/7)
English
this is gonna be fire
himanshu@himanshustwts
i am hosting a conversation on the future of science communication at CAISc by @lossfunk we’ll discuss which forms of science communication still work / which need to be reinvented / what new forms may need to emerge for human-human, human-agent and agent-agent science. register on the luma link below!
English
tensorqt retweetledi
Super excited to have @himanshustwts moderate this discussion. His podcasts on @groundzero_twt, including the episodes with @tensorqt and @Dorialexander, have been some of my favourite discussions to hear on AI!
Who better to discuss Science Communication than a practitioner doing it IRL.
himanshu@himanshustwts
i am hosting a conversation on the future of science communication at CAISc by @lossfunk we’ll discuss which forms of science communication still work / which need to be reinvented / what new forms may need to emerge for human-human, human-agent and agent-agent science. register on the luma link below!
English
tensorqt retweetledi
I wrote this article about @Itsfredi1 with an interview where Manfredi articulated his profound preference for rigid architectural forms as the central language of his artistic practice (rather than relying on figurative or abstract elements).
In particular, Manfredi highlights brutalist structures, which he believes lack any definitive cultural interpretation. So, this interpretive openness compels viewers to engage in a form of self-psychoanalysis, questioning their own identities and perceptual lenses.
Suburbs Gallery@SuburbsGallery
English
this is what the paradigma logo is btw
tensorqt@tensorqt
my phone's gallery producing new logo ideas on its own (these are two separate images)
English
i've been told emanuele will perform tricks and jokes to entertain the attendants, you can't miss it
Emanuele Rodolà@EmanueleRodola
happening next week in Seoul. beers and food are on us, actual taste is open for discussion!
English
tensorqt retweetledi
Today, we are releasing verifiers v1 — an overhaul of our environment stack for the modern era of agentic RL and evals.
We decompose environments into a taskset, a harness, and a runtime.
Run complex agentic tasks like coding and computer use at scale, in any harness.
English
tensorqt retweetledi
excited to share the news! what if and many of my pieces will be exhibited at @SuburbsGallery starting Jul. 13th.
I'll share more info below. Links and further details soon
Suburbs Gallery@SuburbsGallery
gm.
English
i wonder what kind of privileged information they used when training GPT-5.6-Sol with self-distillation starting from GPT-5.5 and not doing anything else
English
tensorqt retweetledi
Today marks the end of my last week at Nvidia.
I joined with the rest of the excellent Gretel team when we were acquired back in April of 2025, which now seems like 10 years ago thanks to the time-dilation of AI progress.
Over that year, I got to do quite a lot:
help build the best synthetic data generation tool in the industry (NeMo Data Designer),
scaling it out for pre & post-training datasets for Nemotron by building some slick cluster tooling,
contributed to 4 (!!) Nemotron LLM builds (Nvidia doesn’t mess around with open models),
took a small merging experiment reproduction from small scale (30B 😅) up to 550B and pass along some pretty significant eval compute savings as a consequence for our pretraining heros,
pull my hair out over the state of public evals & benchmarks,
and most importantly, got to collaborate with some of the best researchers and engineers in the field, all working to advance the frontier of open-source AI and build the best computing platforms in the world to support it.
Thank you to everyone @ Nvidia & Nemotron for welcoming me into Team Green 💚. Thank you to the Gretelers for your support over this journey 💜.
English
insane unlimited in-house brewed craft beer
tensorqt@tensorqt
we're incredibly oversubscribed for this. if we're friends and you'd like to attend lmk now, or i can't guarantee i'll be able to get you in
English
we're incredibly oversubscribed for this. if we're friends and you'd like to attend lmk now, or i can't guarantee i'll be able to get you in
Paradigma@paradigmainc
Novel paradigms require novel conversations. We’re hosting a social in Seoul, during ICML, to gather the best people for those conversations, over a few beers. See you there, link in the first comment.
English
insane stretch of the definition of “neolab”
Alexandr Wang@alexandr_wang
compute daddy @dylan522p has spoken
English
