Robert Lundberg

Life is a game, let's be creative and have fun with it! 🎮

Currently doing ML stuff @RedHat
Previously Co-founded @PerceptiLabs

[Elias Bareinboim @eliasbareinboim]

@fivetroop @ylecun @vkhosla @GaryMarcus @yudapearl There really isn't a mystery or open problem here, in response to Dirk, and complementing
@yudapearl's answer. The question on how to incorporate causal assumptions on neural methods has been studied and implemented in practice, here's an example: https://t.co/Yx5VPyfOEs.

[Robert Lundberg @robert_lundberg]

Back to playing with DQN!
Now can we create a causal graph from the latent space of the model?

[Robert Lundberg @robert_lundberg]

Does a Q-network-based RL model already embed cause and effect?🤔

[Robert Lundberg @robert_lundberg]

Just watched https://t.co/SpiOyI40e6, fantastic video!
Around 39 minutes, @karpathy mentions there is no simple evaluation of whether the model did well.
This may be naive, but can't the LLM give multiple answers to a question and argue against itself which is the best answer?

[Robert Lundberg @robert_lundberg]

@hwchase17 @StabilityAI @databricks @langchain @huggingface Awesome!
Have you come across anything like auto-prompting? Where for example GPT-4 iterates on prompts based on successful and failed examples.

[Robert Lundberg @robert_lundberg]

@jobergum Thanks a lot!
Is there an ETA on what “soon” means (Q1,Q2,…)? :)

[Robert Lundberg @robert_lundberg]

Hey @jobergum, what's the best way to create a nested structure of NN/ANN rankable tensors in Vespa, similar to how you would imagine tightly coupled parent/child documents to work, where you can choose to query only the parents or the children?

[Robert Lundberg @robert_lundberg]

@jobergum @vespaengine Ok great, thanks!

[Robert Lundberg @robert_lundberg]

Hey @jobergum, is there any way to limit hnsw's space based on an attribute (userID where they should not access each other's documents for example) in @vespaengine?

[Robert Lundberg @robert_lundberg]

@jobergum @vespaengine It's both, each user have their own set of documents that should not be accessed by others, I'm just looking for what the best way to set that up would be so I don't get hit by scaling issues.
Thanks for the blog, I've seen it earlier, by brute force you mean exact search?

[Robert Lundberg @robert_lundberg]

@jobergum @vespaengine Yea, I'm looking to do pre-filtering, but worrying a bit about the latency impact at scale (high amount of users, few documents per user).
Correct me if I'm wrong, but it would not be sustainable to have one document type per user, right?
Many thanks for the reply! :)

[Robert Lundberg @robert_lundberg]

@Suhail @DmitryKan has some really nice material for vector databases

[Harrison Chase @hwchase17]

🚨Biggest @langchain drop yet🚨

⚡️Dynamic, zero-shot composition of multiple chains

Easily plug in different subchains (google, REPL, wikipedia, DBs) just by telling the router LLM (in English!) when and how to use them

`pip install langchain==0.0.8`

https://t.co/vxBzH0v5F9

[Robert Lundberg @robert_lundberg]

@Suhail I'm really impressed with how smooth the user experience on playgroundai is! Do you have any tips/tricks for how you set up the infra?

[Robert Lundberg @robert_lundberg]

Are you struggling to validate pains or want to double-check that you are building something important for your users?
Try using Customer Problems Stack Rank (CPSR)!

Great visual thread: https://t.co/Ek9D1wrQ5i

How to execute: https://t.co/i0WblPl7tJ

[Shreyas Doshi @shreyas]

A B2B Product Management Story: on discovering problems that customers actually care about

Very visual story thread👇🏾

[Andy Zeng @andyzengineer]

With multiple foundation models “talking to each other”, we can combine commonsense across domains, to do multimodal tasks like zero-shot video Q&A or image captioning, no finetuning needed.

Socratic Models:
website + code: https://t.co/Zz0kbV5GTQ
paper: https://t.co/NpsW61Ka3s

[Robert Lundberg @robert_lundberg]

Ever had a joke fly over your head? (I know I have)
This AI got your back (∩^-^)⊃━☆ﾟ.*･｡ﾟ

[Karol Hausman @hausman_k]

Me: Nobody understands my sense of humor
AI: I got you.

[Robert Lundberg @robert_lundberg]

@aidangomezzz @_RobToews Fantastic article, thanks for sharing!

[Turing Post @TheTuringPost]

NVIDIA open-source 3 versions of the StyleGAN architecture:
1. The original StyleGAN implementation
2. StyleGAN2
3. StyleGAN3

StyleGANs is a technique for creating photorealistic synthetic images.

Find links here: https://t.co/W6T4OXxO40

[Robert Lundberg @robert_lundberg]

Then after finishing up for the day, I add everything that did not complete (plus new things that came in) back to the list to review tomorrow.

[Robert Lundberg @robert_lundberg]

I start my day 🌅 by writing down everything that's in my "soon to be done" scope.
Then I categorizing them into different buckets "Do now", "Do after lunch" and "Do if time permits / another day".
This way I have a clear list of what to focus on. 🗒️

How do you start yours?