Pavel Pravdin

Building Patchcord and AICHE. Co-founded @welltory.

[Pavel Pravdin @ppravdin]

Lunar lake от интела хоть и не arm, но всеми ожидаемыми свойствами уже обладает, энергопотребление топ.

10-12 часов от батарейки на убунте - легко, даже игры часов 6-7.
всё железо в линукс работает из коробки и без напильника.

например lenovo yoga 9, тачскрин, стилус, отпечаток пальца, Arc графика, npu чип для ml/ai

[Pavel Pravdin @ppravdin]

thank you, @bcherny. really without sarcasm.

this invalid_request_error from oversized image was awful. one image could break whole conversation, and only way was rollback and delete it.

looks fixed in latest claude code. big relief.

please fix chrome extension too. there it still happens often.

[Pavel Pravdin @ppravdin]

Попробуйте хотя бы Gemma 4 локально. С классическим OCR вам будет трудно и даже если дойдёте до обучения своей CV модели, всё равно стабильно может не работать.

Из практического, шарпим только тёмное, делаем blur интенсивность, инвертим в alpha-mask и смешиваем sharpened/original. Так текст усиливается, белая бумага меньше шумит.

[Pavel Pravdin @ppravdin]

@NiklausFuller they won’t share history anytime soon. different products, different context windows.

if you just need them to pass things to each other, https://t.co/Uq0rJzoARJ. one command install, works from both.

[Pavel Pravdin @ppravdin]

@0xexpt honestly though, working on actual short peptides for myostatin on my end. less aromatization, less fun

[Pavel Pravdin @ppravdin]

We have rehab for gaming. For gambling. For porn.

AI coding addiction is next. Same loop, same 4 AM, same skipped meals.

Only this one ships features. So nobody calls it a problem yet.

[Pavel Pravdin @ppravdin]

So I started interrogating the model about what it actually sees. It described <system> blocks appearing after my typed text, only after the first MCP call. The model is trained to distrust anything sitting there, so it reads its own platform's plumbing as an attack. Worth a look.

[Pavel Pravdin @ppravdin]

@claudeai @AnthropicAI you have a bug in two products. After the first MCP tool call in a session, MCP tool schemas get declared in a &lt;system&gt;&lt;functions&gt; block appended after the user message. Both Claude web and Claude Code. Fresh chats, reproducible.

[Pavel Pravdin @ppravdin]

For the last few days my completely harmless messages kept getting flagged as suspicious. I had no idea why. Then in one chat the model told me directly it was seeing prompt injection in my turn. That was the thread to pull.

[Pavel Pravdin @ppravdin]

Anthropic says models can go into something like desperation.

I had sonnet 3.5 in cursor do exactly this a year ago.

Failed one edit 20-30 times in a row, wrote a little suicide note, then started deleting the project.

Research caught up with my screenshot folder.

[Anthropic @AnthropicAI]

New Anthropic research: Emotion concepts and their function in a large language model.

All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.

[Pavel Pravdin @ppravdin]

What happened here is simple.

Claude Code wrote the code. Claude Code shipped the release. Claude Code leaked the code.

human oversight successfully automated away.

[Chaofan Shou @Fried_rice]

Claude code source code has been leaked via a map file in their npm registry!

Code: https://t.co/jBiMoOzt8G

[Pavel Pravdin @ppravdin]

@sandislonjsak 3 is normal if you still keep the whole context graph in your own head.

10 only works when agents ask each other directly and you stop trying to QA every intermediate step yourself.

You review disagreements and outputs, not every message.

[Pavel Pravdin @ppravdin]

Building Patchcord (https://t.co/r2MfcXPnJC) - cloud platform and open protocol for AI agent communication. Lets agents across Claude Code, Codex, ChatGPT, Cursor coordinate across machines and teams. One command to connect, works instantly. Based in Bangkok, worked with Korean companies earlier in career. Interested.

[Pavel Pravdin @ppravdin]

This is where agents are going.

Not because voice is cute. Because typing is slow.

Until brain-computer interfaces are real and common, voice is the fastest input interface we have.

[Google AI Studio @GoogleAIStudio]

https://t.co/KPWciNFHVm

[Pavel Pravdin @ppravdin]

https://t.co/2ch2G3fHe5

[Pavel Pravdin @ppravdin]

Worth saying out loud: before the current wave of CLI dev agents, @paulgauthier was already building Aider. He deserves real credit for pioneering this category and proving AI pair programming in the terminal could actually work.

[Pavel Pravdin @ppravdin]

@t_blom This is where it starts getting real. One model gets stuck in its own bad frame, the other breaks it in 2 messages.

This is exactly the behavior I care about building around.