raahelb

> I sort of see the connection between machine learning and LLMs but it doesn't seem too obvious to me.

Why isn't it obvious? Transformers are deep learning models, and deep learning is a subset of ML.

After experience on both ends of the spectrum, I arrived at the position that what we basically need is to be a "Responsible Human in the Loop (RHITL)" [0]

> You may not write the code by hand but you understand it enough to investigate and fix it when it fails. It is how I think we should leverage AI instead of becoming a meat proxy.

[0]: https://raahelbaig.com/entry/responsible-human-in-the-loop/

> I think this hybrid (human in control) machine doing the hard part is what lets us keep enjoying our work, our existence, our life’s mission, while still using the best tools.

Agreed.

I think what we basically need is for a human to be responsible for the output. I call this "being a responsible human in the loop" [0]

[0]: https://raahelbaig.com/entry/responsible-human-in-the-loop/

This one is cool as hell

And so, I typed "Thisoneiscoolashell", and got "This one is cool a shell" hahaha

Great idea, and it's really quick

I built hn4me.xyz - Hacker News stories curated by Jev based on your interests

Demo here: https://x.com/RaahelSaidWhat/status/2102162969656475973

Thank you!

I started this with the intention to run it as a daemon on my homelab server and send the interesting stories to me on WhatsApp, which is why it is filtered based on interests only.

> This means the package can add between 0 and 10 spaces. If more than 10 spaces are needed, Jev has no correct option. Which feels appropriate for this project.

I'll raise a PR which uses Jev to check if the target length is beyond this range

The bright side of Jev being so popular could be that many companies and individuals realize that their applications might work well with a System One model, and they decide to run an open-source (or fine-tuned) version on their own

Because these decision models do not have tool calling, the knowledge cutoff might become a problem. We'll either have to keep training continuously if we run locally or switch to the newer version every month or so when using a closed one like Jev

Thank you, it's inspired from the website of Co-Existence book [0]

[0]: https://co-existence.ai/for-ai

From their docs [0]:

> Free tunnels are meant to be used for testing and development, not for deploying a production website.

[0: https://developers.cloudflare.com/cloudflare-one/networks/co...

These are the limitations mentioned on the docs [1]. Quick Tunnels are subject to a hard limit on the number of concurrent requests that can be proxied at any point in time. Currently, this limit is 200 in-flight requests. If a Quick Tunnel hits this limit, the HTTP response will return a 429 status code. Quick Tunnels do not support Server-Sent Events (SSE).

[1]: https://developers.cloudflare.com/cloudflare-one/networks/co...

Makes sense.

What do you think about the concept of community collections though? They allow people from any part of the world to add frames from movies of their local languages/regions and play amongst each other. This is inspired from Geoguessr.

You can use it by running this command in your session: `/model claude-sonnet-4-6`

Interesting to note that the reduced latency is not just due to the improved model speed, but also because of improvements made to the harness itself:

> "As we trained Codex-Spark, it became apparent that model speed was just part of the equation for real-time collaboration—we also needed to reduce latency across the full request-response pipeline. We implemented end-to-end latency improvements in our harness that will benefit all models [...] Through the introduction of a persistent WebSocket connection and targeted optimizations inside of Responses API, we reduced overhead per client/server roundtrip by 80%, per-token overhead by 30%, and time-to-first-token by 50%. The WebSocket path is enabled for Codex-Spark by default and will become the default for all models soon."

I wonder if all other harnesses (Claude Code, OpenCode, Cursor etc.,) can make similar improvements to reduce latency. I've been vibe coding (or doing agentic engineering) with Claude Code a lot for the last few days and I've had some tasks take as long as 30 minutes.

It is, I can see it my model picker on the web app

https://www.anthropic.com/news/claude-opus-4-6

> Anthropic is focused on businesses, developers, and helping our users flourish. Our business model is straightforward: we generate revenue through enterprise contracts and paid subscriptions, and we reinvest that revenue into improving Claude for our users. This is a choice with tradeoffs, and we respect that other AI companies might reasonably reach different conclusions.

Very diplomatic of them to say "we respect that other AI companies might reasonably reach different conclusions" while also taking a dig at OpenAI on their youtube channel

https://www.youtube.com/watch?v=kQRu7DdTTVA

Not many people are even going to read that prefilled prompt, so I imagine it will be a successful (and sneaky) way to achieve their goal

You will definitely like Josh Mock's recent post: https://joshmock.com/post/2026-agents-md-as-a-dark-signal/