arbayi

https://web.archive.org/web/20260701201606/https://www.theve...

i think we have the causation backwards here. llms aren't expensive because they have to be — they're expensive because we keep reaching for the expensive model instead of putting any effort into making the cheap one good enough.

a surprisingly large fraction of production workloads can be handled by smaller models with the right scaffolding. it's often easier to switch to a larger model than to engineer those pieces, so many teams never bother.

my intuition is that a lot of the current "ai cost crisis" is really an orchestration problem rather than a model pricing problem. before asking whether frontier pricing is sustainable, i'd first ask how much of that spend is simple tasks being sent to the smartest available model by default.

my bet for the next few years is that the model itself stops being where the value is. frontier models will become more like commodities, and the real difference will be the layer around them as routing each task to the cheapest model that can do it well, verifying the output, and only escalating when needed.

eventually, asking "which model do you use?" will sound a bit like asking "which cpu do you use?" the engine still matters, but the system built around it matters a lot more.

device: https://www.waveshare.com/wiki/ESP32-S3-Touch-AMOLED-2.06

what llms reason best in might not be a language at all. more like a graph. graphs say more in less space than prose, and they stop the model from wandering off.

not the author of the gea but from what I can see in the readme, the ideas go back to 2017, erste.js and regie were earlier versions of the same concept.

https://github.com/dashersw/erste https://github.com/dashersw/regie

not the author of the gea but from what I can see in the readme, the ideas go back to 2017, erste.js and regie were earlier versions of the same concept.

https://github.com/dashersw/erste https://github.com/dashersw/regie

For a fun side project, we built a quick demo to interview our OpenClaw agent called Mahmut. We used LemonSlice, LiveKit, and ElevenLabs. We just shared the full interview, and here's the repo if anyone is interested in building the same thing: https://github.com/openserv-labs/openclaw-voice-avatar

Our avatar was a really hard one to handle, but I think it's still really good. All we did was connect the APIs. LemonSlice is really cool.

I just found about them on twitter and tested, its so crazy that it work so smooth and fast. It's probably powered by Claude Agent SDK in behind, I created a fully functional telegram bot under 3 minutes. It seems perfect for non-technical users.

https://x.com/i/status/2012322509400531005

Working on WordPecker, trying to create something like a personalized Duolingo but for vocabulary you actually encounter.

Started when I was struggling to read books in English. Pushed an open source version back then (https://github.com/baturyilmaz/wordpecker-app), later added more features, and now working on a mobile app.

Recently started testing alpha version, fixing bugs and introducing new features right now (https://alpha.wordpeckerapp.com/).

My end goal is to build: an AI language learning companion that knows what you read, listen, and watch, knows you as a friend (real life, who you are), then helps you improve using that context. If you're B1 at language, it creates a personalized path to get you to B2, then C1, and so forth using your context.

https://wordpeckerapp.com/

We need AI Bowl!!

that’s great to hear. I’d love to check out your project and have a chat if you’re up for it. maybe i can contribute to it instead of working on a separate project.

I want to build a complete radio station managed by an AI agent. It’s still at the idea stage right now: https://github.com/baturyilmaz/agent-radio

The idea is simple, but I think it could be really cool: an autonomous agent that actually manages an entire radio station. It creates its own shows, play copyright-free tracks, shares the daily program schedule on social media and the website, and later I want to add guest appearances too and live 7/24.

We are running 6 different LLMs to do predict the future using polymarket and there's a chairman agent which follows the MOST EDGE from LLMs.

I believe, this is the easiest and most practical AI product to use among all the ones we’ve seen this year.

https://github.com/dashersw/liquid-glass-js maybe you can also check this out?

I’ve even thought about building my own DIY e-ink reader/note-taker for this.

How does apps for Paper Pro works?

From the original linkedin post:

"AI has a product problem. Not a model problem.

Models are making capability leaps every few weeks but AI-native product innovation hasn’t kept up.

Most products are forcing AI into existing UX patterns rather than rethinking an AI-native experience from first principles. Parallels to early mobile (2007-2010) - for years, mobile meant shrinking your website into your phone until Uber reimagined transportation.

To be clear, there are lots of amazing AI-first products. See NotebookLM, Lovable, Stitch, Flow. But the pace of innovation on models has been faster, for a lot of reasons which I will explore in future posts."

https://www.edgeaifoundation.org/livestreams

https://store.vufine.com/ I think vufine might be what you are looking for, I don't know much about them as well, also found it here on Hacker News.

I just found out about them and it seems super good. I wonder why Meta doesn't support such a thing called "Meta Glasses Application Store"

https://github.com/google-ai-edge/gallery

A gallery that showcases on-device ML/GenAI use cases and allows people to try and use models locally.

Hello!

I've been playing around with agents, MCP servers and embedded systems for a while. I was trying to figure out the best way to connect my real-time devices to agents and use them in multi-agent workflows.

At OpenServ, we have an API to interact with agents, so at first I thought I'd just run a specialized web server to talk to the platform. But that had its own problems—mainly memory issues and needing to customize it for each device.

Then we thought, why not just run a regular web server and use it as an agent? The idea is simple, and the implementation is even simpler thanks to MCP. I define my server’s endpoints as tools in the MCP server, and agents (MCP clients) can call them directly.

Even though the initial idea was to work with embedded systems, this can work for any backend.

Just wanted to share and would love to hear your thoughts—especially around connecting agents to real-time devices to collect sensor data or control them in mutlti-agent workflows.

Repository: https://github.com/openserv-labs/mcp-proxy

Repository: https://github.com/enessusan00/openserv-mcp

I've been experimenting with OpenServ, a platform for managing multiple AI agents working together. It lets you create teams from an agent pool to handle specific tasks, and you can build agents via their UI or TypeScript SDK. Each agent is basically an Express app with capabilities as routes, so you can integrate any agentic framework you like. I've built some and shared them on my GitHub.

I’m starting to feel like most agentic frameworks are more theoretical than practical—except for easy-to-implement text embedding capabilities like those in LangChain and AutoGen. That said, I think platforms like OpenServ, which provides SDK while allowing devs to leverage other agents from the same ecosystem, will be the ones to win. I’m also planning to explore other platforms like LangFlow, CrewAI, n8n, etc., to compare different approaches.

OpenServ's demo: https://www.youtube.com/watch?v=l9Vpko9GXRg

Here's a list of all the apps (built by them!) mentioned in comments:

- LangTurbo (by @sebnun) - langturbo.com : Learn through podcasts with transcriptions and contextual word definitions

- Nuenki (by @Alex-Programs) - nuenki.app : Browser extension that translates appropriate-difficulty sentences across websites, with hover-for-definitions feature

- Manabi Reader (by @wahnfrieden) - reader.manabi.io : Japanese-focused integrated reader with SRS and Anki integration

- (by @muth02446) - Spanish: appicenter.net/Apps/VocabES/ - English: appicenter.net/Apps/VocabEN/ : Uses spaced repetition and audio for basic vocabulary learning

- Vocabuo (by @kebsup) - vocabuo.com : Combines SRS flashcards with ebook/YouTube/website reader, using AI for content generation

- LingoStories (by @laurentlb) - github.com/laurentlb/lingostories/ : Open-source language learning tool

- Turkish Learning Tool (by @learning-tr) : Browser extension for colloquial translations with audio and pronunciation features

- Language Reactor (by @davidzweig) : Planning to open-source soon, looking for contributors

Note: above list is summarized by Claude 3.5 Sonnet.

Harika görünüyor! Are you planning to share it open source?

Not: Bu benim en sevdiğim şarkılardan biridir!! (it's one my fav songs)

Hi!!! Thank you for being the 100th* star and for suggestion! I've added this to priority list.

*thanks again!!

Thank you! The app doesn't have such a feature yet, but I think it will be needed eventually, and it has been recommended before. I will add this to the roadmap.

Would be cool to have such a tool, maybe an extension using chrome's built-in AI APIs?