Phi

[Phi @sublimotion]

@bernhardsson I think it depends on the definition of successful.

[Phi @sublimotion]

https://t.co/UMBBfzFtiG

[Phi @sublimotion]

The Agent Verification Spectrum
Claude Sonnet 4.6 passes 51% of SWE-bench. Sounds good until you realize you can't tell which half is correct.
We gave agents verification tools. 69.5% of tool users solved the issue. 18.8% of non-users. Same model, same benchmark.

[Phi @sublimotion]

@karpathy @peteskomoroch Frontmatter and yaml are actually indexes for agents.

[Phi @sublimotion]

@togethercompute Thanks for sharing this important work!
Sharing some related experiments I have done related to the verification gap on those coding agent trajectories.
More details here:
https://t.co/fCXyw6fV5v

https://t.co/WNwvFGSpxJ

[Phi @sublimotion]

@lihanc02 Great analysis and timely.
Sharing some related work I have done related to the verification gap on those coding agent trajectories.
More details here:
https://t.co/fCXyw6fV5v

https://t.co/WNwvFGSpxJ

[Phi @sublimotion]

The bitter lesson has a time horizon problem https://t.co/XYc1qDeWTM

[Phi @sublimotion]

If you happen to be coming to the #Raysummit come and learn about

🚀 Mixture of Experts & Disaggregated Architecture on vLLM & NVIDIA Dynamo

🍻 Come network, grab a bite or a beverage

Register now: https://t.co/Kinw45t9oj

[ray @raydistributed]

Explore 2⃣ scalable approaches to train thousands of models to a geographic location feature in batches w/ Ray Core in record times
1⃣ Distributed reads per task
2⃣ Centralized Ray Object store reads per Task

Checkout the blog + code 👇

#manymodelsml
https://t.co/HtmmC2ROzs

[Robert Nishihara @robertnishihara]

One of our goals with @raydistributed has been to provide a great off-the-shelf experience for beginners as well as the performance and flexibility required by power users. @OpenAI is on the "power users" end of the spectrum.

https://t.co/7nPWlJlQRp

[willem @willpienaar]

Building a feature store on AWS? Check out this post on running Feast on AWS with S3, Redshift, and DynamoDB https://t.co/zk4ZjoNaWV

[Phi @sublimotion]

@bindureddy Would that be called decision science?

[DeepGraphLibrary @DGLGraph]

We released the Drug Repurposing Knowledge Graph (DRKG), a new dataset for finding drugs to fight #COVID19 , a joint effort with UMN, OSU and HunanU. Check out our blogpost about using DGL-KE to train embeddings and rank drugs: https://t.co/LsFGEZXrSp

[Phi @sublimotion]

@AndySeaborne Hello Andy. Is it possible to use Fuseki on top of another triple store? I was looking to run it on top of AWS Neptune. I can't find any example or reference in the documentation of Fuseki running on top of another triple store.

[Greylock Partners @GreylockVC]

"What Keeps CIOs Up At Night" Key learnings from the first meeting with our Tech Leader Advisory Council. https://t.co/YPZolbCyD4

[Phi @sublimotion]

@pvanimpe @sublimotion
I do speak french but based in Boston MA unfortunately.

[Phi @sublimotion]

@pvanimpe @noootsab @xtordoir @bigdata
Hello, I tried to use the discount code but it says it is invalid.

[Mike Loukides @mikeloukides]

Odd way to put it, but good summary of Spark's place: Spark is Too Big to Fail http://t.co/15h2h326yO

[Kirk Borne @KirkDBorne]

Top 5 Reasons to Get Your Graph On: http://t.co/yMudJtP5jk #LinkedData #BigData #Analytics

+ https://t.co/IVbiHZAbJG

[Phi @sublimotion]

Join @Jet_Insider for a 6-mo. membership to a better way to shop + compete for 100k stock options #sharesforshares https://t.co/tcTJbz4KSS