hung

becoming uncommon amongst uncommons

the ui for agents is heading into a bottleneck for innovation. every current tool is missing certain components to reach its full potential. vs code only has seamless ssh mode and git integration. the only reason i open an ide nowadays is to see diff views, git changes, and the built-in terminal in one place. but tmux/ terminal is impractical at scale. if i have 12 agents running in a grid of 12 cells, every time i want to dive deep into a session i need to zoom in and scroll, click references to jump to source code, watch streaming logs of background processes, cross-check different sub-agents. what if i found out some interesting bug that worth investigate more and need to spawn another panel right next to the current one? it's constant friction. claude code and codex cli suffer from same problems while claude code does exceptionally well on making the experience better rapidly. codex app with the hacky ssh mode patch is the closest to the ui i can imagine, but it's far from perfect. i'm skeptical of uis that overload your cognition with noisy information. when you give too much customization without constraints, it becomes more of a hurdle than a feature. on the other end, something like openclaw is a rare candidate that has the most comprehensive context ingestion behind the scenes. Yet its full potential is bottlenecked by basic messaging interfaces like imessage and telegram, which were designed for a completely different purpose. what's missing across all of them: there's no concept of high-level agent management -- managing a team of agents the way a ceo manages an org. and the "just get a bigger monitor" answer only solves the problem indirectly. most of our daily workflow isn't sitting at a desk. we walk, bike, run, switching between environments. we mostly rely on phones and small laptops with our voice as lowest latency form of transmitting information. maybe brain computer interface is the great unhobbling? six months to a year ago, the ui bottleneck was tolerable. but it's getting to the point where every day, every feature added is another push against the ceiling of the current interface paradigm. i'm optimistic that this pressure will drive us to experiment with something fundamentally different, and we're starting to see that emergence, with the right components scattered across different apps, waiting to be pulled together.

To return Americans to the Moon, NASA is shifting to an iterative, execution-focused approach – just as we did during Apollo.  We are standardizing rocket architecture, embedding NASA expertise across industry, and increasing launch cadence to support sustained lunar operations.   We are sending a demand signal for crewed missions beyond Artemis V, with at least two providers capable of bringing astronauts to the surface every 6 months.  The goal is not just to reach the Moon, but to stay.  America will never give up the Moon again.

This is a man who has been haunted since childhood and built a billion dollar company as a side effect of trying to make the haunting stop.

this encapsulates well my review. The book truly ignites my interest in physics and understanding of the universe!

@braelyn_ai fitness sf fillmore and presidio ymca are great

this is the first time I’ve ever wondered where I can find a pool in San Francisco

@lineesh_antony @ValsAI @grok @askvals help me answer atony’s question

@ValsAI @grok which mini max is from which company

Initial results are in for Minimax 2.7, and it comes in at #12 overall on the Vals Index. If the weights are released, it will be #2 on the open-weight index (only 0.5% behind #1).

something i learned from my wife, who recently learned how to sew: do not do beginner projects. if what you want to make is difficult to make, you should just try to make it. don't do a slow learning process. don't start with the basics. start with the advanced

@ValsAI @askvals where is @grok model in the leaderboard?

GPT 5.4 is #1 on Vibe Code Bench at 67.4%, +5.7% higher than the previous SOTA. This is our benchmark that measures model’s ability to produce an entire working application from a short text specification.

@SQMah great work SQ!

Just demoed some of 5.4’s computer use and frontend capabilities - check it out here! What I really like is that computer use was on an Electron app, so Codex can also make and test desktop apps as well Also yes I need a haircut :)

@chribjel GPT-5.4 best but too slow

So whats best for coding now? gpt-5.3-codex or gpt-5.4?

Some details I think people might miss with this release. - What stands out about GPT-5.4 is that the model spends time checking its own work much more than other models - navigating the browser, clicking around, poking at edge cases, running backend scripts to test the DB. The ratio of editing/ verifying is flipped, and I think it’s a paradigm shift. - Coding benchmarks are not saturating! There's still a massive difference between "model can build a static single-page website" and "model can build a fullstack app with deployment, connect to payments and email services, and actually test it before handing it back to you." GPT-5.4 is much closer to the latter camp but still not quite there yet. There are still many more aspects of coding to evaluate that we’re cooking!

time and time again i keep underestimating the power of going out of your way to surround yoursel with the right people.

@scaling01 @askvals how does this compare with your number?

Qwen3.5 27B looks really fucking good on ArtificialAnalysis Leaderboard it's on par with DeepSeek-V3.2 and Minimax-M2.5

@claudeai @askvals have you already run eval for this model?

This is Claude Sonnet 4.6: our most capable Sonnet model yet. It’s a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. It also features a 1M token context window in beta.

@askvals what's best model to vibe code today?

We have been built and used AskVals internally and found it really helpful. Go try it and let us know your thoughts. More features to come!

“I love struggling, actually. It makes me feel alive” – Alysa Liu

cli is the new ui

"claude, simplify the claude make no mistake"