Kaggle
Kaggle is a global community of practitioners, researchers, and enthusiasts building and advancing the frontier of AI. Through AI competitions, benchmarks, and agentic evaluation, Kaggle provides a proving ground for community-led innovation.www.kaggle.com
40
MENTIONS
23
EPISODES
20
PODCASTS
Search complete. 40 mentions across 23 episodes found for "Kaggle".
Sep 13, 2026
Naman PandeyHOST
Step two, post about said cool things.
Naman PandeyHOST
So is that would you would you agree roughly that, you know, just find something that interests you solves a specific problem is not like a Kaggle data set that is just anybody can tell that this was done purely to serve as a bullet on your resume under your project section.
Naman PandeyHOST
Is that roughly the mindset to think through with this in terms of what to post?
Shrey ShahGUEST
Yeah, I think I'm glad that you asked this question.
speaker_4NARRATOR
The model selects data, prompts, methods, reinforcement learning techniques, and evaluation feedback loops within a time limit, without training directly on held-out evaluation data.
speaker_4NARRATOR
MLE Bench revised tests, data science and machine learning performance on Kaggle competitions involving designing, building and training models on GPUs.
Zvi MowshowitzHOST
All tests showed substantial improvement in results.
Zvi MowshowitzHOST
None were earth-shattering.
NeilHOST
Meta is flexing some serious autonomous research muscle today, too.
NeilHOST
Their autonomous research system just competed fiercely on Kaggle.
AlexHOST
For those unaware, Kaggle is a massive data science platform.
AlexHOST
Companies post real-world problems and offer cash prizes for solutions.
AlexHOST
It is a brutal proving ground for data scientists.
NeilHOST
Meta's system beat almost 4,000 human competitors on real tasks.
NeilHOST
It actually finished in the top 10 overall.
AlexHOST
People don't realize why Kaggle is so incredibly difficult.
speaker_0HOST
Meta's research credentials back it up too.
speaker_0HOST
Their Era three agent won gold in NVIDIA's Kaggle challenge, placing eighth out of roughly four thousand teams.
speaker_0HOST
The takeaway for anyone selling agents to businesses, consumer agents are maturing fast, so client expectations are about to jump.
speaker_0HOST
On the quieter end, iOS twenty-seven adds spatial reframing in Extend for Photos and upgraded cleanup for object removal.
André CebedaGUEST
But something that's a little more intensive is going into our open source tech stack.
André CebedaGUEST
So using stuff like Kaggle to store your data, Python, of course, to work on data automation.
André CebedaGUEST
accessing an LLM and then through natural language processing, just ask an application of questions to get geographic answers that you want.
André CebedaGUEST
So that's building and programming an application that wouldn't use any of SRE's stack.
Cliff WeitzmanGUEST
You could be a machine.
Cliff WeitzmanGUEST
And so we hire a lot of math Olympiads and LeetCoders, Kaggle award winners, and people who, like, studied physics and math, like they might've even not coded before.
Cliff WeitzmanGUEST
Because I just need the hunger and the work ethic and the intelligence.
Cliff WeitzmanGUEST
Anyone can become so good so fast now.
TomHOST
That dynamic will inevitably accelerate defensive initiatives across the hyperscalers.
TomHOST
Expect an immediate push to build out independent, federated model registries, whether that means Google aggressively ramping Kaggle and Vertex Model Garden or open-source consortiums establishing decentralized, non-commercial weight mirrors specifically engineered to prevent single-vendor choke points.
TomHOST
There's also a massive cultural inflection here.
TomHOST
Hugging Face built an authentic identity as the collaborative counterweight to closed proprietary APIs.
David HolzmüllerGUEST
MLPs will still need a GPU, but they do scale to quite large datasets.
David HolzmüllerGUEST
Like we're also seeing on Kaggle, like they have these tabular playground competitions now that for the last maybe year or so have been basically always datasets that have like 600,000 samples.
David HolzmüllerGUEST
And there is always like MLPs or booster trees at the top.
Ravid Shwartz-ZivHOST
But not like, uh, like, um, I don't know, TabFM or something like that.
David HolzmüllerGUEST
I mean, TabFM
Ravid Shwartz-ZivHOST
is
David HolzmüllerGUEST
probably also hard to run on Kaggle GPUs, I would imagine, but maybe it gets close nowadays.
Allen RoushHOST
What about, um, what about things like the loss functions when they're, they're training and what about like it's cross validation still being used?
Stephen FollowsGUEST
But I can write whatever formulas we need to build like, okay, I want all that by year, by genre.
Stephen FollowsGUEST
And then as for getting them, there are lots of established data sets online, whether it's Kaggle or just searching for stuff.
Stephen FollowsGUEST
I find that quite often there is a kind of quasi...
Stephen FollowsGUEST
So there's full data sets that have permissions and whatever.
Anca PetreHOST
La deuxième chose, c'est un concours.
Anca PetreHOST
Cette même année, le laboratoire Merck met en ligne un défi ouvert sur la plateforme Kaggle.
Anca PetreHOST
Le problème posé est un problème de QSAR, l'héritage direct de Hansch qu'on a vu au tout début de l'épisode, prédire l'activité de molécules.
Anca PetreHOST
N'importe qui sur Terre peut participer.
13 more episodes mention Kaggle.
Create an account to see the whole feed, search across every transcript, and follow the entities you care about.