Jake Leventhal (@JakeLeventhal) header
Jake Leventhal (@JakeLeventhal) profile picture

CTO building @RiotIQ and Artelo

Florida, USA
Joined May 2021
Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)Photo by Jake Leventhal (@JakeLeventhal)
Jake Leventhal retweeted
I came to realize people liked the french #slopcore without having the lyrics 😅 sorry! So here is a version with english subtitles!
274
638
313
6,044
1,108,903
Jake Leventhal retweeted
v2.5.13 is out! Another patch with many bug fixes! Special thanks to @jakeleventhal for the great help in improving type-inference lint rules. biomejs.dev/internals/change…
3
45
2,498
Jake Leventhal retweeted
Your IQ Guaranteed Within 4 Points: Take Riot's Digital Test Today 🎯

3
7
23
131,266
Jake Leventhal retweeted
AI;DR (AI; didn’t read)
95
2,444
89
18,093
439,067
AI getting more intelligent does not mean that software teams will automatically get exponential productivity gains. To truly benefit from increased model capabilities, teams need to create systems that leverage increased intelligence to scale their output (skills, loops, etc.).
2
89
Jake Leventhal retweeted
🟦⭐🟦⭐🟦🟥🟥🟥🟥🟥🟥 ⭐🟦⭐🟦⭐⬜⬜⬜⬜⬜⬜ 🟦⭐🟦⭐🟦🟥🟥🟥🟥🟥🟥 ⭐🟦⭐🟦⭐⬜⬜⬜⬜⬜⬜ 🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥 ⬜⬜⬜⬜⬜⬜⬜⬜⬜⬜⬜ 🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥 ⬜⬜⬜⬜⬜⬜⬜⬜⬜⬜⬜ 🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥
43
31
4
1,138
67,744
openai training sol be like
I see what they are doing here lol. The tuned 5.6 sol's max test time compute to achieve just below mythos but at vastly superior token efficiency. Look at that its a straight line, thy could blow way past it if they inference scaled it.
11
49
3
1,448
68,826
Jake Leventhal retweeted
No model survives the "are you sure?" They all fold instantly.
414
634
124
26,451
1,288,330
Imagine a company makes $650M/yr. The CEO signs a 4 year deal on key equipment costing $53M/year + $167M guaranteed. Many say it's too risky from past breakdowns. He doesn't listen. 1.5 years in, he exits the deal early, still owing $153M over the next two years. Pic unrelated
1
163
Jake Leventhal retweeted
my wife watching me pull up codex on my phone mid dinner
Codex in the ChatGPT mobile app!
77
268
17
6,711
273,042
The new @linear diff feature is really compelling. One thing problematic about it though is I still find myself needing the IDE to jump around the code (cmd+click, search, etc.). It's critical to the review process for larger diffs. Curious to see if that's what's next.
1
118
Jake Leventhal retweeted
Do you people understand the social rules, taboos and moral principles our society operates under aren’t real right? Whenever they’re inconvenient for the Left or the managerial machine they don’t apply them. They’re solely used as a cudgel against their opponents. Social rules don’t really exist either since people bend social rules around those have the highest status. High status people do things and then it magically becomes “socially acceptable.” Practically none of the ideas of intellectual concepts we believe are actually ground in logic or fact or historic precedent. It’s just stuff we made up on the spot in the 20th century. So why do people live and act like they’re in a cage? Constantly wondering what others think or what’s ok to do. The reality that others cannot accept is that other people largely don’t care enough about you to police your behavior. When they do so it’s in the most petty and banal way possible like how you dress or your vocal tone that you can just modify by context. Everyone in our society is too numb to actually understand this or enforce the system. I can’t wait for people to realize this since we won’t have to live with this bullshit anymore. I’m not saying I want anarchy. I’m saying that we’re hobbling ourselves in a thousand little ways we don’t need to and if we ripped the bandaid off we could actually live good lives and fix things, not this screen mediated purgatory.
49
59
7
446
23,060
Hey @shadcn are there any examples in documentation of mutli-app (different themes) monorepos? I have a setup like this where each app extends a base set of components and overrides the classes to apply different themes. This allows for apps to share common UI elements. The new preset/theme guides are starting to become incompatible with my workflow. Any ideas for where I can look for guidance?
5
29
11,177
This is cool and all but tab completions stopped being relevant a while back. Agents, worktrees, and diff/git tools are where tooling is needed, and Zed seems to be stuck in the past. I want Zed to be good but it's missing so many core features that it feels like a beta product.
Zeta2 is here. 30% better acceptance rate than Zeta1. 200x more training data, LSP-powered context, faster predictions, open weights. Try it now in Zed. We didn't just improve the model. We rebuilt the entire data pipeline behind it: zed.dev/blog/zeta2
1
171
Rust may be fast but everyone shipping rust hardly ships at all
1
74
Jake Leventhal retweeted
Codex app/T3 Code/Conductor should copy the file panel from Cursor Glass New agentic IDEs still need a file view, even if a simplified one. Opening another app for this is too much friction. Cursor nailed the ux imo
52
8
7
510
70,592
Replying to @ericzakariasson
@ericzakariasson @leerob Some first impressions/feedback on Glass: Overall it's quite refreshing and removes a ton of the VS Code noise but it's missing a few key features. 1. Worktrees. This is now my go to method of starting tasks in Conductor that makes starting local work much easier and lower friction. 2. Git integration is very minimal. The most important feature for me in an editor is reviewing diff. I use Cursor instead of Conductor for viewing diff simply because of VS Code's superior diff viewing (character diffs, staged vs unstaged, etc.) and for the Git Graph extension (notably this landed in @zeddotdev recently). Glass currently seems to only let me stage entire files, not hunks, or specific lines, which makes it feel very vibe codey in its current state and not for serious work. 3. Threads within workspaces. This is something that Conductor crushes and it makes it super easy to do multiple things at once within the same task. Need this feature. 4. Pull request reviews. I currently exclusively review PRs via Cursor with the GitHub extension. If Glass doesn't add this I cannot use Glass. Thanks for reading if you do, also CC @t3code @t3dotchat since the same feedback applies.
4
2
88
I get the sense that maybe the point was to abstract the code away more as agents do more of the work, but until the agent is doing 100% of the work and the review process becomes completely unnecessary because agent code is basically perfect, more review and control is needed.
1
40
In fact, I would even say that reviewing of changes is probably the primary role in SWE these days. Kicking off prompts isn't really the main focus.
37
Jake Leventhal retweeted
sent this to the team today everything great comes from being able to delay gratification for as long as possible and it feels like we're collectively losing our ability to do that
252
696
189
6,853
985,682