Dexter @dexcompiler

courage & leverage dexcompiler.com Amsterdam, The Netherlands Joined September 2022

Tweets

1K
Followers

168
Following

1K
Likes

4K

Dexter @dexcompiler

3 days ago

@OpenAI codex desktop & cli meet all my needs personally terminated my cursor sub moons ago

0 0 4 1K 0

View Details

some papers [while they may still be canonical in their field] are just downright unpleasant to read and it's really unfortunate because a genuinely beautiful set of ideas and concepts could not find a better means of expression and delivery to humanity

0 0 1 21 0

View Details

Dexter @dexcompiler

a week ago

"recent" non-ai tools and tech I really like: – Cloudflare Workers – Tailscale

1 0 0 35 0

View Details

Dexter @dexcompiler

2 weeks ago

@yongfook not too far from what an ambitious agent would do

0 0 0 32 0

View Details

Dexter @dexcompiler

2 weeks ago

One of the key problems with the current level of model intelligence is that a model can become quite proficient at satisfying the visible proxy while becoming more dangerous to the underlying system

0 0 0 20 0

View Details

Ben {chmark} Adams ⟠ @ben_a_adams

2 weeks ago

I just proposed an official @leanprover 4 formal semantics for C# - giving C# a foundation for machine-checkable proofs. AI agents are getting very good at writing code - and very good at breaking it. In the agentic era, "tests passed" won't be enough for critical software. We need to be able to prove properties of the code we ship. And this is becoming practical. Rust already has Aeneas -> Lean. Microsoft Research is using it to formally verify SymCrypt. Java and C have roots into formal verification too. C# should not be left behind. AI can write the code. AI can construct the proof. A deterministic proof kernel checks whether the proof is actually correct. That is a very different trust model from: AI writes code -> AI reviews code -> hopefully both were right. Proposal: github.com/dotnet/csharpl…

10 28 182 11K 51

View Details

Dexter @dexcompiler

2 weeks ago

@alliekmiller If you start with the premise that they're def not running what the public is running in terms of infra and model harness then it shouldn't be surprising at all Btw context window issues are almost disappearing from the timeline

0 0 0 286 0

View Details

Dexter @dexcompiler

2 weeks ago

@anah_sahh Agency And maybe one small addition: taste

Andrej Karpathy @karpathy

2 years ago

Agency > Intelligence I had this intuitively wrong for decades, I think due to a pervasive cultural veneration of intelligence, various entertainment/media, obsession with IQ etc. Agency is significantly more powerful and significantly more scarce. Are you hiring for agency? Are

2K 9K 50K 11.5M 37K

0 0 1 326 0

View Details

Dexter @dexcompiler

3 weeks ago

@mitchellh Because AI simply has no taste [or has poor taste] I go through the pain of building a full design system on which my frontends are based every time I need a good frontend Some parts of said design system are now compressed into a skill but the work still needs to be done

0 0 0 65 0

View Details

Dexter @dexcompiler

4 weeks ago

The challenge though is that this way of working in the current era can be really annoying but only when this mode switching is not considered part of a broader strategy to keep you tethered

0 0 0 13 0

View Details

Dexter @dexcompiler

4 weeks ago

To stay "deep" interested folks might need to run a barbell strategy on their AI work modes: have some projects where you directly type out stuff without agents [while going thru the slow, now-excruciating back and forth with some llm] AND have another set of projects where it's all agents with you running point [if you like]

1 0 2 36 0

View Details

Dexter @dexcompiler

4 weeks ago

@thdxr It's weird that folks would even think it's anything more than demand management

0 0 0 283 0

View Details

Dexter @dexcompiler

4 weeks ago

@joshm How about: maybe this one's not for public consumption? And it's just sufficient to function as revolutionary leverage for those who know how to wield it

0 0 0 425 0

View Details

Dexter @dexcompiler

4 weeks ago

@theo Nah not really The unfortunate ones are those who have zero compass sense of how to engineer solutions to problems

0 0 60 7K 0

View Details

Dexter @dexcompiler

4 weeks ago

@scaling01 like humans agents probably hate long meetings I dont see the problem we made them

0 0 1 225 0

View Details

Erik Meijer @headinthebox

4 weeks ago

"If a Millennium Problem falls next, it will likely be P vs. NP (via a breakthrough in theoretical computer science/quantum complexity) or the Navier-Stokes existence and smoothness equations. The Riemann Hypothesis may take slightly longer, as it requires highly sophisticated leaps in analytic number theory, though AI agents are already accelerating formal progress around the Prime Number Theorem." Gemini

15 8 105 24K 46

View Details

Dexter @dexcompiler

4 weeks ago

@diegohaz Just read this again and it sounds like I've been stoned since 5.6 launch fml

0 0 0 9 0

View Details

Dexter @dexcompiler

4 weeks ago

@diegohaz You dont need to use Sol Ultra. Ive been xhigh since launch without any problems. But maybe my harness since 5.5 represent my alpha

1 0 1 30 0

View Details

Vaibhav (VB) Srivastav @reach_vb

4 weeks ago

YES! Codex can quickly check in whenever it needs a decision, even outside plan mode, so it doesn’t run with the wrong assumption add to ~/.codex/config.toml: [features] default_mode_request_user_input = true

Ben Davis @davis7

4 weeks ago

Add this to ur Codex config, makes it so that Codex will use the multiple choice tool outside plan mode. Makes it feel 100x better

87 200 4K 565K 6K

36 72 1K 136K 2K

View Details

Dexter @dexcompiler

a month ago

the weird human brain i have tends to think some of the names of these codex subagents will have the traits of these mythological and factual characters :)