ubeyt

husband, building things for humans and machines, sharing what I learn along the way.

[ubeyt @ubeytdemir]

calling it now, jensen comes back. guy had too much aura to just disappear 🤷

[Mario Nawfal @MarioNawfal]

🇺🇸 🇨🇳 Jensen Huang agreed to buy Hugging Face for $12.9 billion in September... and that one move tells you what he believed and why he had to go.

Philip Pilkington says Hugging Face was America's answer to DeepSeek, open source with no one controlling the pipeline, and Huang bought it because his whole model depends on open source AI: make the best chips, let the software be free, and beat China on hardware.

Closed source is the opposite model, with Anthropic and OpenAI charging $20 a month forever in what Philip calls rent extraction.

That is the model the gray suits are now protecting, and the man who threatened it most just resigned.

@philippilk

[Jun Song @jun_song]

If open source loses, our future will very likely be a dystopia.

[ubeyt @ubeytdemir]

@_mohansolo any plans for steering mid run or an incremental mode? also open to tips on using antigravity properly, maybe im holding it wrong 😅

[ubeyt @ubeytdemir]

got gemini ultra to test it properly in @antigravity. cheap, yes. but a few things keep killing it for me 🧵

[ubeyt @ubeytdemir]

codex does small incremental changes, you see them one by one. gemini tries to one shot the whole thing. incremental wins for me, easier to catch it early and steer

[ben hylak @benhylak]

if you're vibe-coding mobile apps, just do native swift.

models are so much worse at writing performant expo apps right now.

[ubeyt @ubeytdemir]

3.Humans are complex. Your brain processes emotions and facial expressions faster than AI ever can thanks to evolution. We call that "feeling." And that feeling right now is disgusting.

Look at how this avatar looks at the candidate. Yes, the candidate's behaviour is absurd, but a real person would laugh at it.

[ubeyt @ubeytdemir]

Hiring is hard, and it looks solvable by AI. However, the problem isn't tech; there are a few reasons why AI interview apps are failing 🧵 :

[Defiant L’s @DefiantLs]

I guess this is a thing now... evaluation is conducted or scored by AI rather than a human recruiter

Many job seekers and commentators describe AI interviews as cold and dehumanizing with some dropping out rather than interacting solely with a machine

[ubeyt @ubeytdemir]

2.Humans aren't just numbers, and great headhunters can see more in a person than a prompt can.

[ubeyt @ubeytdemir]

@fuxsocy_py @0xTib3rius I mean mitigation step is blocking the post owner :D

[ubeyt @ubeytdemir]

@fuxsocy_py @0xTib3rius Blocking this guy

[ubeyt @ubeytdemir]

@dhh except the BASIC prompt never burned $40 of tokens on a typo 😅

[ubeyt @ubeytdemir]

karpathy says ~20 years for a 1B model that actually thinks. way sooner for sure

[Codez @0xCodez]

Andrej Karpathy:

“A 1B model can be much more effective than most current LLMs.

90% of current models waste a ton of capacity and money on remembering stuff that doesn’t matter.”

here is his 5 core thesis about Small LLM’s:

1. the intelligence was never the knowledge. Strip the memorized internet, keep the reasoning. What’s left is the cognitive core, and it’s small.

2. the core might fit in ~1B params. Maybe less. “Plenty of room at the bottom.” We’ve been paying for a library when we needed a brain.

3. size stopped being the bottleneck. What makes an agent work is the tools it can call and how tight the feedback loop is, not parameter count.

4. intelligence lives in the system, not the weights. A raw LLM is a CPU with no OS. Small core + good harness > giant model alone.

5. you can own it. A model on your hardware is one no one can rate-limit, price-hike, or deprecate on you.

in 12 months the gap between teams shipping on-device and teams paying per-token frontier bills won’t be capability.

TwIL an 8.8B model at 90%+ of frontier isn’t a benchmark flex.

It’s the thesis Karpathy’s been repeating all year, made real.

study more on this below ↓

[ubeyt @ubeytdemir]

@NousResearch @WhiteNightNiki Terrible reply indeed

[ubeyt @ubeytdemir]

@PovilasKorop depends on the use case tbh. gemini argon was surprisingly strong on swift for me