Scott Condron

3K posts

Scott Condron banner
Scott Condron

Scott Condron

@_ScottCondron

Helping build AI/ML dev tools at @weights_biases. I post about machine learning, data visualisation, software tools.

Dublin, Ireland 参加日 Nisan 2018
2K フォロー中5.7K フォロワー
固定されたツイート
Scott Condron
Scott Condron@_ScottCondron·
Here's an animation of a @PyTorch DataLoader. It turns your dataset into a shuffled, batched tensors iterator. (This is my first animation using @manim_community, the community fork of @3blue1brown's manim) Here's a little summary of the different parts for those curious: 1/5
English
34
490
2.6K
0
Scott Condron
Scott Condron@_ScottCondron·
.@wandb LEET (TUI) is art 🖼️ Now with console logs and system metrics
Scott Condron tweet media
English
5
5
63
5.8K
Scott H. Hawley
Scott H. Hawley@drscotthawley·
@_ScottCondron OMG, I missed the announcement! Downloading this app immediately. Looking forward to not having to fiddle with mobile browser.
English
1
0
1
54
Scott Condron がリツイート
Weights & Biases
Weights & Biases@wandb·
Robotics AI evaluation is uniquely hard. Your models perceive, reason, and act in the physical world. Outputs are videos, trajectories, sim results, not just loss curves. We shipped tools in W&B Models built for exactly this! 👇
Weights & Biases tweet media
English
4
9
126
241.6K
Scott Condron
Scott Condron@_ScottCondron·
The future of agent UIs
English
1
0
5
306
Scott Condron がリツイート
Morgan McGuire
Morgan McGuire@morgymcg·
h/t to wandb eng for a few nice autoresearch-focussed tweaks to the scatter plot
Morgan McGuire tweet media
English
1
3
6
1.5K
Mariusz Kurman
Mariusz Kurman@mkurman88·
Wandb truly needs a mobile app
English
3
0
4
149
stochasm
stochasm@stochasticchasm·
Yearning for a nice wandb mobile app
English
15
1
57
3.5K
Scott Condron がリツイート
Weights & Biases
Weights & Biases@wandb·
We heard you. The wandb mobile app is now LIVE on iOS 🚀 Monitor training runs from anywhere. Crash alerts the second something breaks. Live metrics on your phone. This has been the most requested feature in wandb history and it's finally here!
English
13
31
280
586.3K
Scott Condron
Scott Condron@_ScottCondron·
A nice pattern for autonomous research: Start from a baseline model LLMs generate ideas Fan out coding agents in parallel Evaluate each branch Merge the best changes Repeat @wandb is already great for tracking all of this: artifacts, traces, diffs, evals, and lineage
English
1
1
11
1.6K
Scott Condron
Scott Condron@_ScottCondron·
@minu_who I’d like to give both hanks a whirl, would you mind sharing :)
English
1
0
1
15
minu
minu@minu_who·
So I dumped the entire output folder from Prior Arts hank into this one, and asked it to make it pretty: I'll leave you to watch the video for the output: youtube.com/watch?v=QNg-xr… There was no human in the loop, except for the one who sealed a shard of their taste (or even trauma) into these hanks. If you want to try either hank, lmk! Happy to :)
YouTube video
YouTube
English
3
0
6
355
minu
minu@minu_who·
After months of scepticism, I stopped doing my literature review manually. Two hanks wrote it instead, then made it pretty. No human in the using of the hanks, only in the making.
English
3
4
22
4.6K
Sriraam
Sriraam@27upon2·
I need an IDE for RL research. Who’s making this? I need a tool that can help me understand existing research, look at their datasets, understand their evaluation metrics, compare insights across papers, and also for my own experiments. I’m vibe coding UIs for each paper and environment, but I want an agent to have reusable primitives that compose well across modalities and task domains, do a preliminary lit review, show me just the parts that I care about without me having to find the repo, download the dataset and prompt Are there tools ppl use @willccbb?
English
9
2
85
9.7K
Nicholas Bardy
Nicholas Bardy@NicholasBardy·
@_ScottCondron Is it public? I had emailed shawn a version of this I've been doing myself or almost a year now. works very well on top of wandb api to compare runs and pupeteer to screenshot charts
English
1
0
0
35
Scott Condron
Scott Condron@_ScottCondron·
Research repositories for auto-research agents is essentially MLOps + LLMOps observability - tracked experiment configs & metrics What was tried, why, what worked, what didn’t - versioned files What files changed - e.g. training code for neural architecture search - agent traces What was the agents reasoning, tool calls, model/prompts used Some open questions: - how the research context is exposed to the agent e.g. scratchpads of experiments, convenience CLIs to navigate through prior work - the agent harness / runtime e.g. where is it running, sandboxed, resumable etc. I think you want this all to be very flexible. The types of experiments it’s doing, how expensive the evaluation runs are, and how you want to navigate and debug/learn/reproduce it’s work will vary a lot
English
4
1
21
1.9K
Scott Condron がリツイート
Yinjie Wang
Yinjie Wang@YinjieW2024·
OpenClaw-RL Technical Report! Make your🦞@openclaw stronger by just using it. We propose a method that combines the advantages of GRPO and OPD, and evalution results. The repo is already 1.7k stars now, feel free to contribute! Come in and have fun~ @MengdiWang10 @LingYang_PU
Yinjie Wang tweet media
English
36
123
704
58.6K
Scott Condron がリツイート
Morgan
Morgan@morganlinton·
The cofounder and CTO of Perplexity, @denisyarats just said internally at Perplexity they’re moving away from MCPs and instead using APIs and CLIs 👀
Morgan tweet media
English
329
378
5.1K
2.8M

ディスカバー

@wandb @drscotthawley @mkurman88 @latkins @altryne @stochasticchasm @minu_who @elonmusk