Conor Murphy

[Conor Murphy @cnrmurphy]

I've been working on this little project (clanker assisted) to use local TTS models to convert articles (or any text really) to audio and expose them as an RSS feed so you can listen to them with a podcast app. Kind of nice when I have some reading queued up, but I'm out and about. Check it out - https://t.co/Q42pu3HYSx

[Conor Murphy @cnrmurphy]

@BilwarAmaan Yes, the case and key caps are 3D printed. Then I used some magnetic phone stands as tenting kits.

[Conor Murphy @cnrmurphy]

Anthropic have unveiled their new headquaters

[Conor Murphy @cnrmurphy]

I just saw that the React Compiler was ported to Rust and it made me realize that I'm still waiting for that damn blog post from Jarred

[Conor Murphy @cnrmurphy]

Yeah, so I pretty much feel the same way. I enjoy coding and don't want my skills to degrade, but I realize I would be less concerned with these things if we had access to good local models. So, I think my main gripe is that I don't feel in control of the process in which I've previously found freedom.

[Conor Murphy @cnrmurphy]

@samlambert It's a lot of fun to write code by hand

[Conor Murphy @cnrmurphy]

@thorstenball AGI achieved

[Conor Murphy @cnrmurphy]

@thdxr it's called extremeprompting in corporate world

[Conor Murphy @cnrmurphy]

@theo Every frontier model is a good model until it isn’t

[Conor Murphy @cnrmurphy]

@samlambert ring buffers are definitely the solution here. just make sure they are cache-aligned or your ngmi

[Conor Murphy @cnrmurphy]

@mitsuhiko Wow, a true visionary

[Conor Murphy @cnrmurphy]

So with the whole “agentic loops” thing - I haven’t seen actual implementation details and so far proponents have talked about it in a vague or opaque way - how do you monitor the internals of it? If one agent produces a “negative” artifact, how do you prevent it from poisoning the rest of the loop and future outputs? Like I imagine you need each agent to operate in a very simple and confined way to achieve as deterministic of an output as possible, but I keep imagining one bad artifact posing the “loop” and causing major headaches in course correcting.

[Conor Murphy @cnrmurphy]

@odysseus0z I’m surprised I haven’t seen Norbert Wiener’s name pop up more often after the whole Ralph loop thing

[Conor Murphy @cnrmurphy]

@ThePrimeagen Maybe you are holding it wrong? Instead of describing your problem, try writing loops

[Conor Murphy @cnrmurphy]

Generally, agree - a capable local model is highly attractive for many reasons which I think would draw users away from frontier providers.

Specialized models are interesting especially if it optimizes compute. But it also introduces the problem of fragmentation. There will need to be a good way to orchestrate so many different kinds of models. One good thing about the frontier "generalist" models is that you sort of interact with them the same way for each task you do.

[Conor Murphy @cnrmurphy]

My main motivation was that I didn't want to pay for an app that did this and wanted to experiment with local models (it's more fun to own these things). You do need to run setup Tailscale and run a local server though - so there is a bit of setup involved. But it's worked well for me so far.

[Conor Murphy @cnrmurphy]

I've been working on this little project (clanker assisted) to use local TTS models to convert articles (or any text really) to audio and expose them as an RSS feed so you can listen to them with a podcast app. Kind of nice when I have some reading queued up, but I'm out and about. Check it out - https://t.co/Q42pu3HYSx

[Conor Murphy @cnrmurphy]

@atmoio You’re saying the quiet part out loud bro

[Conor Murphy @cnrmurphy]

I kind of just enjoy running for the sake of...running. Not everything needs to be gamified and maybe you should be reflecting more on your mindset if you need something like this. If you are this reliant on technology, I would encourage you to try to run or at least walk for an extended amount of time without any use of technology - just put the phone and earphones away.

[Stijn Spanhove @stspanho]

🏃‍♂️ I've gamified my own run so I can race my own ghost with the Meta Ray-Ban Display.

I built a web app for the glasses, loaded a previous GPX from Strava, and dropped game mechanics on top.

Pick up coins when you keep pace, sprint zones reward extra points if you push, and a mini leaderboard on the lens shows how you're tracking against your past self in real time.

Best part: it actually works. Seeing your ghost 20 m ahead is a way stronger nudge than any number on a watch. 😅