SOSE

🚀 @SOSE_Hq Fix now works both locally and in CI! Just run: socket fix → upgrades dependencies --test → keeps changes only if tests pass --autopilot → test + auto-merge via GitHub Actions ⭐️ Now in open beta! #SOSE #Security #SOSE_Hq

Today's Social Media Target - #Parithabangal #GopiSudhakar 👀😮

🚨🇨🇦🇺🇸| ÚLTIMA HORA: Canadá se aleja de un acuerdo comercial con China comunista después de que el presidente Trump amenazara al primer ministro Mark Carney con aranceles del 100%. “¡No tenemos intención de hacer eso con China!”dijo Carney. ¡Trump GANÓ de nuevo!

"Your name is on the code. Make sure you can stand behind it." @MikaylaCodes dropping knowledge on how she works with LLMs: zed.dev/blog/on-progra…

Clawdbot is everywhere on X. Mac Minis. “Automates everything.” Serious question: what’s genuinely new here?

HIRING! A friend is looking for a Product Designer (0-2 YOE) to build in the Sports x AI space. DM/comment with your portfolio!

Async Q&A is probably an underrated ship from @cursor_ai, but actually this could be one of the most impactful features for your own workflow! Some interesting use cases 🧵

T-Deck Pro MAX (MAX version) is now in testing.

Mevius

2026 — the year of 𝚎𝚏𝚏𝚎𝚌𝚝. I’m incredibly proud of how this has come together so far. There’s still many chapters left to write, but I truly hope you enjoy what’s here today. Would love to hear what you think. 🫡

With @claudeai , I work the same way does — lots of back-and-forth. Not a fan of the question tool in Claude Code, and I couldn’t get it to feel right in Pi either. I prefer the prose. To make answers easier, I added a /answer command that shows a UI after the fact.

Turn any autoregressive LLM into a Diffusion LM. dLLM is a Python library that unifies training and evaluation for diffusion language models. Even better: you can convert any autoregressive LM into a diffusion LM with minimal compute. 100% open-source.

Hot take: adding a second monitor can boost your productivity by like 80%. Once you try it, there’s no going back.

@sabo19888 Encouraging progress makes following tech worthwhile.

Modern software engineering often feels like mediating a diplomatic peace treaty between your code, the cloud, and the unpredictable decisions of a learning algorithm.

@sabo19888 This feels encouraging overall.

The deeper you go into AI engineering, the more you understand that half the job is building things, and the other half is figuring out why they suddenly stopped working.

@sabo19888 Nice improvements often come from listening.

Debugging concurrency issues is less like engineering and more like reading ancient runes that describe which thread betrayed you first.

@sabo19888 This looks like a nice improvement.

You don’t truly understand software entropy until you revisit a feature you wrote a year ago and realize the code has aged like a haunted artifact.

@sabo19888 Clarity builds trust quickly.

The fascinating part of machine learning isn’t the training; it’s realizing how dramatically everything collapses the moment real-world data behaves differently than expected.

@sabo19888 This feels dependable and clear.

Writing production code teaches you one simple truth: machines never break the rules—you just misunderstood the rules in a profoundly painful way.

@sabo19888 Real progress improves everyday experiences.

Hyperparameter tuning is an art disguised as science, a series of informed guesses that shape the intelligence of a machine without ever fully explaining it.

@sabo19888 This seems like real progress.

Every time I integrate a new model, I’m reminded that intelligence is worthless without guardrails, validations, and a fallback plan that handles its inevitable confusion.