Shelley retweeted
While I really like prototypes, because you can convey ideas easier by interacting with the real thing, it’s also important to know when not to do them.
Prototypes have this flaw that because everything is already built, and it looks so real, people might assume that all the decisions have already been made.
That’s because the building step was the most expensive before AI, but it’s not anymore.
So sometimes, it’s better to ask the why first. Why are we making this in the first place?
Allow yourself to be slow, write a doc with your reasoning, and only after you are aligned, start prototyping.
It all comes down to judgement and taste at the end of the day, whether it’s an animation or an important product decision.
You need to develop great taste, it’s more important than ever.
Shelley retweeted
Investing heavily in post-training at @harvey.
@gabepereyra, @oneill_c, and @mudithj on what we’re working on including 100M-token data rooms and synthetic legal matters for model training
When @mudithj and I met @gabepereyra, we were expecting just another vanilla intro call and instead had the best yarn about research, the state of LLMs, and where intelligence is actually heading. It's rare to meet a founder this deep in the weeds who's also building for one of the most important verticals in this new age of intelligence
So it was awesome to sit down with Gabe for an extended discussion on what it take to build agents that can reliably complete work over hours, days, or even longer?
We talked about why agents today struggle with search and long context windows and how techniques like KV-cache compaction, synthetic data, and continual learning could help.
0:00 Introduction
0:36 Getting legal agents to review the whole data room
2:08 Data rooms larger than any context window
5:28 How far open-source models can go
7:58 Where specialist models fit in legal AI
10:59 Training legal models when client data is off-limits
13:06 Teaching a model how a law firm works
13:59 What belongs in context vs. model weights
15:36 From firm-wide AI to a model for every lawyer
18:37 What training adds beyond retrieving the right cases
20:26 Why context windows have plateaued
24:01 How models could learn continuously on the job
26:12 Can AI recursively improve AI research?
27:07 Research agents can run experiments but not choose them
30:00 Why open-ended research is hard to train
33:47 Why deployment, not intelligence, is the bottleneck
35:08 The cost of frontier intelligence
36:59 Different neolabs, different paths to intelligence
39:26 Using open datasets to compare research methods
41:13 Conclusion
Shelley retweeted
A few photos from design night with @harvey and @DecagonAI. Thank you to everyone who stopped by ✨
Shelley retweeted
Most designers dream of working for Apple or similar. They want to join a team at the top. And I get it, I do.
Maybe its the way I was raised, but I don't see the attractiveness in joining a winning team, quite the opposite.
Especially as a designer, I'd rather work somewhere or on something that hasn't figured out design yet, where you can figure out all the problems and move something, have a real impact.
I see no personal challenge (or fulfilling feeling) in joining a winning team where the bed is already made. Give me the unsexy thing nobody else wanted, give me something messy.
I hope and wish more designers will see it this way in the future too, after all isn't that why we become designers in the first place?
Shelley retweeted
RIP Peter Aldington, whose Turn End is a lovely example of perhaps my favourite architectural style — Kettle’s Yard Modernism.
Shelley retweeted
New Canvas 001 happened last night. @melodyskim and I hosted 60 designers in NYC and almost every speaker showed something personal, not client work, just the tool they wanted to exist for themselves—something that would have felt too expensive or out of reach a year ago.
@jameygannon with s-tier midjourney technique built over constant trial and error, @seanxthielen and @gabrielvaldivia both building personal software for an audience of one, and @neogeomancer on how AI is finally giving form to ideas he's been thinking about for years but couldn't previously express in software.
ironically, the more time we spend building with AI the more I think in-person is going to matter—we're moving faster but spending more time on screens, not less. last night was a reminder that there's no substitute for being in a room with people who care about the same things.
C.S. Lewis asked in 1939 whether beauty matters when the world is in crisis. last night felt like the answer.
Shelley retweeted
The more I see companies talk about coding with agents and designing in code, the worse their UX seems to be.
Users don’t care how it was built or how fast it was built. They just want it to work well.
So walk the walk and make something great.
Shelley retweeted
Dieter Rams, the legendary German designer behind Braun’s iconic products shaped Apple’s design philosophy with “Less but better” approach
Jony Ive, Apple’s longtime designer took heavy inspiration from Rams minimalist aesthetic and his principles for good design
Shelley retweeted
The brand new app icon for @baseapp. Designed in collaboration with the one and only @gavmn.