sbrother

That isn't a mistake AI would make though -- it's something a human would accidentally do when replacing em dashes with hyphens

But it was like this pre-AI, as well. I don't think I spent a significant amount of the work week coding since I was an L3 15 years ago. And even then it was clear that impact (and thus promos and raises) was inversely correlated to the amount of the week spent coding.

What I'm seeing on my teams is everyone being more overworked than ever because the coding part happens SO quickly now that there's no friction to slow down all the nonstop "team alignment and box drawing" work that someone else in this thread alluded to.

Yeah - that is the issue. Doing it that way is going to nuke your reliability rating and you will have to cut prices drastically to get rented.

What I do is… rent on the same platform when I actually need to use a card. A benefit is that if I need a burst of more power, that’s not an issue since it’s available from other hosts. But obviously there’s an inflection point of first party utilization where it makes more sense to own.

You do need fairly high end workstations to put them in, adequate cooling, power and very solid business internet as well. I've put more work into keeping them online than I had expected. But overall the economics are fantastic for it right now, the big question is how long it will continue.

I bought them direct from PNY the week the RTX Pro 6000 was released. Prices are almost double now what they were then.

Yes, they have been near 100% utilization on vast.ai at $1 to $1.25 depending on demand for the past two months. They're in two separate workstations, which I purchased for around $12k each before GPU and RAM prices went nuts this year.

Now, I have absolutely no clue how long this situation is going to last! But the economics don't really work out for local models while it does.

Ok, but I have two of those cards plugged into fiber internet, and they consistently rent out for $1500+/mo on vast.ai. So your example would be $3000, minus a couple hundred for business internet and electricity… that still buys a LOT of $200/mo frontier model subscriptions.

I'm not selling anything; I'm developing it live and there is a landing page, but there's no payment hooked up since like you said, it's not functional yet.

Have you had any luck fine tuning one with musical data for classification or music-aware QA? I've been hacking on https://trebel.la/ which I would like to be a music practice companion, and the biggest missing feature is actually useful audio-based feedback pipeline.

My current approach, not yet validated, is trying to generate training data from masterclass recordings on Youtube, and then fine tuning MOSS-audio on a bunch of those. But I'm interested if there are better models, or large training sets I don't know about.

I really wish I had studied piano for that reason. I put many years into flute and almost went to conservatory for it, so I can play pretty much any melody by ear after hearing it, but I have zero ability to do the same for harmonies/chord progressions. I've tried to learn some piano (and like the parent said, was forced to as part of music curriculum at college) but it's incredibly difficult to learn a new instrument to a satisfying level as an adult.

Yahoo! Search | Engineering Managers, Machine Learning Engineers (Senior, Staff) | US, Canada Remote | Full-time

Our group at Yahoo! Search runs the platform that powers AI chat across Yahoo properties (Finance, Sports, News, Search, etc.) as well as our new AI search surface https://scout.yahoo.com/. Lots of fun problems to work on building a high quality grounding data search engine, agentic chat, and our internal knowledge graph -- all at scale with real users.

We are a fully distributed team but work on US time zones. Working environment is somewhere between big tech and startup world, so experience with both would be ideal.

If you're interested, send me your resume at sam.brotherton [at] yahooinc [dot] com

Yeah, this is going to be interesting. First thing I did with the new Instinct bot was have it generate a property tax value appeal for me. Which will almost definitely go through; it's a slam dunk case but I'd been putting it off for an embarrassingly long time. Now it took me 10 minutes on a Sunday. This is obviously good, but I don't know how public agencies are going to deal with the influx of people trying to get the benefits they are owed, now that they can't rely on red tape and busy schedules to decrease the load.

I have a related question as someone who's been fully remote for about 12 years now, including all of my time in more senior/management roles: how does in person work even work for people who are in meetings all day every day?

My role (eng director) consists basically of 50-60 half hour meetings a week, back to back, with people all over the country/world. It wouldn't be tenable in an open office and it wouldn't even be tenable if I had to physically move across a campus in between meetings. Did people just work/manage across fewer projects and teams?

Great question. I’m using MOSS-audio for that which feels good (and is fast/small enough) but not sufficient to give useful feedback out of the box. I’m working on fine tuning it on a bunch of masterclasses on YouTube and recordings sourced/labeled by my friends. That’s really the blocker for me to feel this is ready for prime time. But the rest of the app already helping me structure my practice.

I'm working on Trebella [0], a tool to help classical music students structure their practice sessions more effectively. I tried to build something similar ~10 years ago in the form of a more interactive music game, but felt like it wasn't actually that helpful for intermediate and advanced students. So this time I'm doing it in collaboration with actual music teachers and building a tool that actually helps me learn. I'd say 90% of my flute practice is done using it at this point, and I'm in the process of taking on the first arms length users in a closed beta. Please shoot me an email if you are interested in any way!

[0] https://trebel.la/

I think they are lying. We all know Sam Altman is a scheming liar; it's not inconceivable that HF is in on this one.

> choose for the ergonomics first

This is unfortunately why I keep buying the Razer Deathadder Pro. It fits my hand perfectly and is super accurate. I hate their software, and the company, but the performance and ergonomics of the mouse are worth it to me.

In theory I could see really enjoying them in an action sports (backcountry skiing, mountain biking, rock climbing) setting. I'd want 1) true AR that could annotate terrain with stuff like slope angle, aspect etc, 2) all the GPS and monitoring functionality of a Garmin watch, and 3) a high quality action camera that could replace a gopro with less faff.

From what I can tell we aren't particularly close to putting this all together in a consumer usable package.

Why did they discontinue that? It was a very good product; we got them for all the grandparents and they worked really well at bringing the family together across distance. Could have fit in so well with WhatsApp too. But then they just killed it.

I saw those and got excited; I've been using Mathematica for decades and do a lot of work with symbolic music.

Unfortunately they are extremely surface level in this release. It looks like there isn't even any ability to load/export MusicXML, kind of weird low level primitives and zero interesting higher level functions. Hopefully they keep iterating on it but I don't think it'll be useful for my workflow right now.

I strongly agree on that being the release where these tools got good enough to substantially speed up my professional work. I have to admit I was super skeptical of AI coding until then.

Oh hey, I'm building something loosely related to this too. Can I ask what need isn't being met by say, Forscore?

I was about to say "what, we've had records for decades" but then I read the changelog.

Interesting. I wonder if there a world where Elixir starts compiling maps to "native records"?

I went all in on the Nest ecosystem when I bought my house eight years ago, and Google absolutely ruined it with the botched acquisition. Half the stuff is Google branded, half is Nest branded, a different half has Google branded software and a different half has Nest branded software. None of it really works reliably anymore. The lock to my front door is completely incompatible with modern "Google Home" and I'm unable to change its passcode.

It's a total disaster and I will never buy Google hardware again.

Love their SaaS offerings though!

Oh dude, I love this. I've been working on an interactive music thingy (https://trebel.la, it's sort of gamified but more designed to structure practice sessions for serious classical musicians) and struggling with the ABC vs. MusicXML choice.

Like most people in the space I'm using ABC for LLM generation (e.g. generating sightreading exercises and etudes) but MusicXML for processing and rendering the output. Would be nice to have something somewhere in between the over-simplified ABC and overly verbose MusicXML.

> …but as a user, I would much rather wait an extra 200ms for my slow/expensive prompt to be accurate

I disagree with this SO strongly. I find the conversational voice mode to be a game changer because you can actually have an almost normal conversation with it. I'd be thrilled if they could shave off another 50-100ms of latency, and I might stop using it if they added 200ms. If I want deep research I'll use text and carefully compose my prompt; when I'm out and about I want to have a conversation with the Star Trek computer.

Interestingly I'm involved with a related effort at a different tech company and when I voiced this opinion it was clear that there was plenty of disagreement. This still surprises me since it seems so obvious to me that conversational fluidity is the number one most important feature.

Same. Well the biggest thing I don't want to do that they could help with is work. But in the cases where it can do that for me, there's no world where that benefit goes to me rather than my employer.

That actually sounds... really nice. Like they've borrowed some of the things from traditional theatre that make the experience not awful.

My reference point for movie theatres is the late 90s/early 2000s so maybe I'll have to give them another try.

wait... I don't think I've ever experienced assigned seats in a movie theatre. Is that a thing?

Maybe I'm missing something as I am not a frontend developer, but when you embed SVGs in an img tag as part of a Phoenix LiveView or even just a static component, you no longer get the ability to dynamically change paths/fills/colors with events coming from the server. Even if it's as simple as having a shape that you want to fill with a brand/highlight color, which at least for me is a common use case.