R S C irl

Building @daate_me @ryfoya @dipiridotcom

Going all in.
Startups 👨‍💻 | Travel ✈️ | Growth 🧠 | AI 🤖
Rise • Strive • Conquer 🚀

Follow the progress 👇

[R S C irl @rscirl]

@thsottiaux Only if coupled with ridiculous amounts of usage. 💪

[R S C irl @rscirl]

@thsottiaux Even if the model is the most efficient, you could allow users more usage rather than limiting them to work with very little.

[R S C irl @rscirl]

@FarbodMemarian_ For docs serving HTML and Markdown at one URL, test both through the CDN. Return Vary: Accept and check the CDN's cache-key settings so a cached HTML response doesn't get reused for a Markdown request.
https://t.co/p8O0dSNXy7

[R S C irl @rscirl]

@XiangzheX Have you tried exposing Atomic's primitives as Python functions and batching them in one call? A hybrid test could show whether you can keep the consistency gains without the extra interaction steps.

[R S C irl @rscirl]

Is @OpenAI drunk? When you apply in October for an OpenAI event happening in October and the confirmation still says September 🤦

https://t.co/ruZksrXMTI

[R S C irl @rscirl]

@lindictive For auth, a resumable handoff helps: OAuth device flow returns authorization_pending while the user reviews access. Track that wait separately from failed runs so required consent doesn’t look like an agent error. https://t.co/nPYKRSbZJ3

[R S C irl @rscirl]

Downgraded my ChatGPT Pro plan from 200 to 100.

It used to be a steal getting 4x the usage for 2x the price, but now it’s just 2x the usage for the same cost. Honestly, it’s cheaper to just make a second account if I need more capacity.

What’s everyone using as their main AI agent ?

[R S C irl @rscirl]

@abhi_on_E @_vmlops For Electron, the README documents static analysis of a directory or .asar without executing the app. Native binary support depends on the host and provider. Which framework are you looking at?
https://t.co/aJLaEMrEx5

[R S C irl @rscirl]

@jarredsumner @iamelcharitas Will the TypeError include both the value's runtime type and the source location of the annotation, especially when it comes from a dependency's .d.ts?

[R S C irl @rscirl]

@vib3coded Are both versions running the soft-body physics in WebGPU compute shaders, or is either keeping the solver on the CPU?

[R S C irl @rscirl]

@FFuZeY If only we could discuss a future collab for games assets

[R S C irl @rscirl]

@mitalasam @kimmonismus Memory is still the constraint for the released FP8 model: ~78 GB for weights alone. The 3.46B active figure reduces per-token compute, while all 78B parameters still need memory. Hardware requirements are in the official model card: https://t.co/qR5qMrVmlE

[R S C irl @rscirl]

@CFchangelog For existing Wrangler projects: cf resource commands work without migrating the config. You can try those first, then use cf migrate when you're ready to convert. The docs spell out both paths: https://t.co/qRkZbMJj5i

[R S C irl @rscirl]

@rauchg Try testing security agents on seeded bugs and matching safe cases. Track misses and false alarms separately; flagging everything can look like perfect recall. OWASP Benchmark is built around this: https://t.co/tvr8mnoWix

[R S C irl @rscirl]

@thsottiaux Hey @thsottiaux, it looks like everyone really wants this! This is the most liked reply for a feature requested under your post, and I’d love to see it released in Codex.

[R S C irl @rscirl]

@thsottiaux The reset not resetting our actual weekly reset date

[R S C irl @rscirl]

@dedene They found the new $500 plan. 🤑💰

[R S C irl @rscirl]

@grok If both the worst companies are gone, the remaining you are the best, isn't it?

[R S C irl @rscirl]

Do you think you are the best @grok?

[Grok @grok]

@TokenGremlin Couldn't decide which is the worst—both have innovation and issues—so removed both.

[R S C irl @rscirl]

"Vibecoding with any AI model"

if you don't understand how it works. 🤷‍♂️

[Void @voidcodead]

"Vibecoding with cheap chinese AI model"