Dmitriy Vasilyuk

Exploring the mindspace continuum.
AI Lead @LittlebirdAI

[Dmitriy Vasilyuk @ReasonMeThis]

no means no, opus. i already miss you, fable

[Dmitriy Vasilyuk @ReasonMeThis]

Word. There’s just so much less you have in to spell out.

[Dhaval singh @Dhavalsingh7]

Combining your screen context with the tools you use is a no brainer and you can see the diff almost immediately vs apps that use just integrations.

Go try it out! ✨️

[Littlebird @LittlebirdAI]

Integrations are here.

Littlebird now helps you work across all your tools and remembers everything.

What you need to do, when you need it, and what that vague task from three Tuesdays ago was actually about.

Need to make a Notion doc? Ask Littlebird.
Need to update your tasks? Ask Littlebird.
Need to inform the team? Well, you know the drill.

[Dmitriy Vasilyuk @ReasonMeThis]

@anieasyy @rasbt Mostly shared, but occasionally I add something to my personal global context

[Dmitriy Vasilyuk @ReasonMeThis]

Codex is having an identity crisis today with "gpt-image-2 does not exist" errors for all requests.

Fix I found: fork the chat. Both the new and original chats started working.

[Dmitriy Vasilyuk @ReasonMeThis]

The "the one honest caveat" is the new "delve". I'd rather have "delve" back.

[Dmitriy Vasilyuk @ReasonMeThis]

@nic_carter The problem here is two terms are being conflated:
- "generally intelligent": LLMs do have _general_ intelligence
- "AGI": this traditionally hasn't been understood literally, but more like "can replace a human on any consequential cognitive task". LLMs are clearly not there yet.

[Dmitriy Vasilyuk @ReasonMeThis]

@simonlast I guess this means you are not too bothered by things like this (and the many other slopifications):

[Dmitriy Vasilyuk @ReasonMeThis]

this is the best explanation of RL:

[Dmitriy Vasilyuk @ReasonMeThis]

@GrugBrainedDev for context: https://t.co/tozSToivJk

there's already /caveman skill for it by @julius_brussee: https://t.co/13PJbojwbm

[Dmitriy Vasilyuk @ReasonMeThis]

hard predict even 5 year ahead but mark my word - in 30 year we all talk like @GrugBrainedDev

first llms and devs (more token-efficient = faster, cheaper), then youngs in texts, then youngs in rl, then everyone in rl.

if i'm wrong eat hat - but nobody will call on it in 30 year

[Dmitriy Vasilyuk @ReasonMeThis]

One of the biggest causes of AI slop in coding agents is that they don't understand or respect contracts.

It's not all their fault. Sometimes it's next to impossible to tell from the code what the contract is.

It would really help though if they used a basic type checker.

[Littlebird @LittlebirdAI]

Introducing Hummingbird: a quicker way to navigate your workflow, uninterrupted.

Built to help you use Littlebird faster and easier across meetings and docs to browsing, research, and deep work.

In an instant, Hummingbird hovers on top of your screen with enough understanding to pull up anything you need, right where you are.

Try it out today.

[Dmitriy Vasilyuk @ReasonMeThis]

@Dhavalsingh7 @cursor_ai Just have Gary the Figure robot push it along

[Dmitriy Vasilyuk @ReasonMeThis]

@ClaudeDevs Wait, only a 4% reduction for a 20k token prompt - why?I would have guessed it would be much larger.

[Dmitriy Vasilyuk @ReasonMeThis]

P.S. it's $80 because I have two cats (for those who astutely spotted the screenshot said $40)

[Dmitriy Vasilyuk @ReasonMeThis]

@LittlebiradAI just saved me $80/month - would pay for a subscription 4x over (if I was paying 😀). Other AIs might have found this too, but the thing is - I asked a different question, it just saw the opportunity.

[Dmitriy Vasilyuk @ReasonMeThis]

The most important step is to admit you have a problem. Now comes the easy part.

[Dmitriy Vasilyuk @ReasonMeThis]

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.