Dmitriy Vasilyuk

Dear @AnthropicAI, @OpenAI, etc., pleeeeeeeease, when you RL LLMs for coding make type-checking first-class in the reward mechanics. Then the constant typing-related messes like in the BEFORE pic could become the pretty code in the AFTER pic. Pretty please.

A quick example of what it looks like when you talk to an AI agent that knows you. No need to explain myself beyond the generic "dry food for cats" - @LittlebirdAI already knows my two boys' names and dietary needs😻

@lifeof_jer I disagree with people who were like "it's all your fault". Everybody messed up there. It's a tiny example of what folks who are worried about AGI misalignment are concerned with on a much more global scale.

So i've been meaning to start being more active on X and one thing I've learnt on how to do this better is to be consitent with posting. 1/N Here's how I used @LittlebirdAI to build my posting schedule, get daily tweet ideas (based on what I'm actually working on), and more.

poker should be taught in school because it teaches you two fundamental life lessons: - how to manage uncertainty - how to live with loss

We are in the uncanny valley era for SWE. young devs never learned to code and experienced devs skills have atrophied. most cannot pass fizzbuzz level questions. Who else is seeing this?

TIL: animals keep evolving into crabs. Very topical. #openclaw youtu.be/_qTPWw9sXmo?si… via @YouTube

Me reviewing

Had an interview with a candidate coming up but couldn't tell who it was from their email in Google Meet. I ask @LittlebirdAI ofc, it goes full detective mode, digs up the exact right resume: first 2 letters of last name + first 3 of first name = email.

Yes, with just one request, if you apply for our backend position, please def ensure_you_actually_know_python

@karpathy We should really start testing all of our ideas that way. People usually hate it when somebody else challenges their long-held beliefs, but if it's an LLM, it's much easier to stomach. Do it, it's good for you!

- Drafted a blog post - Used an LLM to meticulously improve the argument over 4 hours. - Wow, feeling great, it’s so convincing! - Fun idea let’s ask it to argue the opposite. - LLM demolishes the entire argument and convinces me that the opposite is in fact true. - lol The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be careful with the sycophancy.

@LittlebirdAI P.S. The video is called "Everything We Got Wrong About Research-Plan-Implement", really worth watching for anyone who codes! youtube.com/watch?v=YwZR6t…

Just a small example, but pretty telling - it's not obvious to some people what to do with @LittlebirdAI once they download it. Honestly, you don't even need to do anything special, just use it like Claude/ChatGPT/etc - you'll get your aha moment when you least expect it!

Tried to find the video I saw of @dexhorthy talking about his new agentic coding learnings to share w/ my team. 3 mins of fruitless search thru my history, I give up: go to @LittlebirdAI, dictate a semi-coherent description of the video. Boom - 20 sec later I have the link!

We're live on @ProductHunt today. If Littlebird has saved you time, helped you stay on top of things, or just made your day a little easier, we'd really appreciate your support.

@alexframegreen Was awesome to wake up to this trending - after staying up super late improving Littlebird's self-knowledge. If you download it, try asking it about its own capabilities, it should be quite "self-aware". I think you'll love it like we do!

Checked - and of course nothing is new under the sun, something akin to this has been done before, for example: - arxiv.org/pdf/2306.05426 - arxiv.org/pdf/2409.14586

My contribution to LLM innovation (I haven't checked if anyone has suggested this already): LLMs should have a "backspace" token that allows them to go back and redo part of their response if they realize they went in the wrong direction. You are welcome! :)