@inprealpha

/s

Joined January 2020
  • Tweets141
  • Following261
  • Followers5
15Media
hi @theo some UI frictions in t3 code for me. t3 code seems to have a lot of multi function ui elements which makes it quite difficult to quickly parse and build muscle memory (esp where hotkeys aren't available). for example the search bar is also used to name a project here.
11
this is so useful. had a mod running to do exactly this (and some more stuff). @amorriscode does this work with the desktop app as well? claude says it doesn't.
Lydia Hallie ✨
@lydiahallieOct 14
We recently added idle compaction to Claude Code! It now compacts long idle sessions right before the prompt cache expires (~1h). If you come back to it later, it only has to reprocess the summary instead of the entire conversation, which saves you some usage.
14
Auto-continue is huge! Anyone else would like if there was an option for the session to auto-compact right before reaching the usage limits and then auto-continuing when done in @t3dotcodes?
1
7
altho ofc compaction = deletion so this should be an option (threadwise? / projectwise?)
7
LinkedIn
🤖 Made with AI
Tibo
@thsottiauxSep 10
Replying to @nudevise
Shouldn’t have underestimated Astra
20
People who are complaining about ChatGPT work. Do not understand that no one outside of our circle understands what an agent is, and they do not care.
30
gpt 5.6 sol computer use means I never have to do tedious stuff on my computer
31
"bring it all together" is such a powerful phrase when speaking to llms
34
Apparently, they are holding off on this for now!
ClaudeDevs
@ClaudeDevsMay 13
Starting June 15, paid Claude plans can claim a dedicated monthly credit for programmatic usage. The credit covers usage of: - Claude Agent SDK - claude -p - Claude Code GitHub Actions - Third-party apps built on the Agent SDK
48
wait do people really think that anthropic employees who are non-u.s. citizen will be locked out of mythos? so many things they could do: - newer checkpoint - maybe they have a larger model - or any other internal version basically infinite ways to get around it.
34
Wake up guys another eval dropped. As an aside, I like evals which disagree with the popular narrative.
Cognition
@cognitionJun 8
Introducing FrontierCode: a coding eval that raises the bar for difficulty & quality. Each task took 40+ hrs of work by leading open-source maintainers. Models write sloppy code that works but isn’t maintainable. Our eval is first to measure: would you actually merge this code?
44
As much as I love gpt-5.5 for coding or even research task, Trey is 100% spot on.
Trey Goff
@thetreygoffJun 5
I don’t know how to put into words why Claude Opus is so much better than GPT So I’ll try to explain with a bunch of examples instead:
1
49
The madmen did it. Holy grail of agent eval!
Arena.ai
@arenaJun 4
Introducing Agent Arena: real-world agentic evals at scale. How do you evaluate agents doing actual work? We measure millions of live sessions where real users accomplish real tasks. On Arena, models now get web search, filesystem, and terminal tools to complete complex workflows: writing code, creating slide deck, researching the web, building apps, and analyzing documents. Every session produces rich signals. Users iterate with the agent turn-by-turn: approving, editing, correcting, praise or expressing frustration. The environment gives feedback too: shell errors, tool failures, recovery attempts, and more. Our leaderboard measures each model's agentic performance using causal inference across five signals: task success, steerability, error recovery, user praise vs. complaint, and tool hallucination. This leaderboard snapshot is built from 300K+ tasks, 2M+ tool calls, and 40M lines of code by agents. Top labs in Agent Arena: - #1 @OpenAI: GPT-5.5 (High) - #2 @AnthropicAI: Claude-Opus-4.7 (Thinking) - #3 @Zai_org: GLM-5.1 - #4 @GoogleDeepMind: Gemini-3.1-Pro - #5 @Kimi_Moonshot: Kimi-K2.6 More analysis in the thread, with the full technical blog below.
1
2
9
1,936
Seriously, though, this is really the only way to do a truly general purpose eval for agents. The only fundamental limitation differentiating this from regular use I can think of is that this eval only scores one short results.
1
36
Oh I think @daradoescode you use Windows too right?
13
you know the worst part about the claude code changes? there is now no gui support for linux to use with the sub
24
what is up with everyone on x and ruining a decent take by exaggerating everything up to fucking heavens
21
new reverse engineering post when @she_llac ???
Boris Cherny
@bchernyApr 16
Opus 4.7 uses more thinking tokens, so we've increased rate limits for all subscribers to make up for it. Enjoy!
46
never change gemini
43
Seen a decent amount of people saying how good Cursor's Harness and product is these days, and complaining about how expensive it gets without the OpenAI and Anthropic subsidies. Composer 2 could be a game changer if these gains hold up in real life. cursor.com/blog/composer-2
1
53
Load more