doh · HackerTrans

30 giorni fadiscuss

[flagged]

I built a very similar server myself [0] with a similar setup. I run different models for different purposes, but the primary one currently is kimi 2.6. I run kimi as the orchestrator model and then qwen, Gemma and others for specific tasks (sometimes loaded dynamically based on the task at hand), all exposed through the pi harness. I also use Hermes for some personal repeated tasks which connects to the same models, hosted on my local Mac Studio.

I am not even going to pretend that this is financially reasonable option. I simply wanted to have a local models. Maybe down the line, as cloud models become less subsidized, I might benefit from having a local setup, but for now, it wasn't the most prudent financial decision.

But one big benefit is that I never have worry about my account being randomly banned nor I have to worry about running out of quota. I still use codex and opus for some specific tasks, but as tools are improving, I need them less and less.

[0] https://x.com/synopsi/status/2024235558193811778?s=20

doh

2 mesi fadiscuss

I have built something like this and in process of collecting the data.

Frontier users: 527,865 Light indexed: 527,865 Ready to queue: 9,083 Fast scores ready: 0 Activity events 24h: 30,266 Fast scores completed 24h: 19,123 Deep jobs completed 24h: 3,043 Fast-score ETA: n/a Deep-hydrate ETA: 69h Stale running jobs: 0 GitHub backpressure jobs: 19,113 High automation signals: 4,608 Medium automation signals: 1,327 Completed jobs: 74,714

Biggest challenge is Github's rate limits. At this pace it will take two more months to have 98% coverage. But after that the maintenance should be quite straight forward.

doh

3 mesi fadiscuss

I don't want to defend LLM written code, but this is true regardless if code is written by a person or a machine. There are engineers that will put the time to learn and optimize their code for performance and focus on security and there are others that won't. That has nothing to do with AI writing code. There is a reason why most software is so buggy and all software has identified security vulnerabilities, regardless of who wrote it.

I remember how website security was before frameworks like Django and ROR added default security features. I think we will see something similar with coding agents, that just will run skills/checks/mcps/... that focus have performance, security, resource management, ... built in.

I have done this myself. For all apps I build I have linters, static code analyzers, etc running at the end of each session. It's cheapest default in a very strict mode. Cleans up most of the obvious stuff almost for free.

doh

3 anni fadiscuss

I think that’s the crux of the point. Americans are trained to be demanding customers. Whatever is then packaged as a product/service for purchase, will be treated equally.

As you pointed out, students pay a lot of money for their education in US. Shouldn’t they have a say in it? If that’s not good for the society, maybe it shouldn’t be sold as a product in the first place.

doh

3 anni fadiscuss

As a European living in the US it took me many years to adjust to the return system in here.

In "Europe" (I'm super generalizing here), to return a product, it has to be in pristine condition, you have to the original receipt and even then they will go over it like you are smuggling drugs into the country. In US, nobody cares. This is built into the business model. You pay higher prices due to returns, but then you buy more stuff because you don't worry about returns.

When I first moved to the US, I was extremely careful with money (didn't have any). So every purchase was carefully planned and I worried about it long after it was done. Then I made a mistake and bought pair of shoes. They were uncomfortable. I felt like an idiot. But I already wore them, so now I am stuck. I told my colleague about it and he looked perplexed.

After bunch of back and forth he decided to teach me a lesson. He took me a to a grocery store. Bought a can of corn. We walked out, he opened the can, dumped half of it in a trash can. We walked back to the return counter. My heart was pumping so hard I was losing sight at moments. He put it down on the counter and said: I would like to be refunded. I thought the clerk is going to smack him silly and me with him. But he didn't even flinch. Gave him few coins and we were on the way.

This was 18 years ago. I since very well adjusted to the system, maybe a little bit too well.

doh

5 anni fadiscuss

Pex | Multiple positions | REMOTE (US and EU), FULL-TIME | https://pex.com/careers/

Pex is building Stripe for digital rights. Our Attribution Engine is being deployed on most of the UGC platforms enabling any creator to be paid for their content and be able to freely mix and remix already copyrighted content without worrying about takedowns. We recently announced $57M round to help us bring our services to the masses (https://pex.com/blog/pex-announces-57m-in-new-funding/).

We are hiring for numerous positions, including:

  - Lead Front-End Engineers
  - Senior Front-End Engineers
  - Data Engineers & Data Architects
  - Machine Learning Engineers & Researchers
  - Site Reliability Engineers
  - Senior Designers
  - Reverse Engineers (assembly)

For all employees, we offer:

  - equal salaries regardless where you live
  - 30 days of paid time off
  - day off on your birthday
  - fully covered health benefits [US] (gold/platinum) [70% coverage for dependents]
  - 4 months paid parental leave
  - balanced work/life (no weekends, late nights, extra long days, ...)

If you want to learn more, read about our hiring process (https://pex.com/blog/interviewing-at-pex-heres-what-you-can-...) and on our culture (https://pex.com/blog/pex-culture-focusing-on-what-really-mat...). You can also reach out to us at [email protected]

doh

5 anni fadiscuss

Search engines are generally considered fair use [0], at least in US. So up until HN doesn't send cease and desist, it's in the clear. Once it does, then it's all about what the claims are. Google litigated the crap out of this [1] over the years, so there is a positive precedent minefield all over the place.

[0] https://www.everycrsreport.com/reports/RL33810.html

[1] https://en.wikipedia.org/wiki/Fair_use#Text_and_data_mining

doh

7 anni fadiscuss

> Maybe the fix is to require the submission to include the timestamp start and end.

This is already a requirement [0] that was introduced couple of weeks back.

[0] https://support.google.com/youtube/thread/9566717?hl=en

doh

7 anni fadiscuss

> Do I see systems like Content ID going away without a change in law? Absolutely not.

Considering the new EU Copyright Directive, especially with Article 17, if anything, the world is moving closer towards CID-like systems than away from it.

doh

7 anni fadiscuss

For the curious ones, this policy change impacts slightly less than 10% of all potential claims (not claims that were made, but could've been made).

Quite large portion of these potential claims are short videos that are being re-uploaded from other platforms, like TikTok and Instagram.

Disclaimer: I run a company that monitors YouTube and other platforms for analytical purposes, some related to copyright. Our system is able to identify segments as short as 0.5 second which allows us to produce this kind of information.

doh

7 anni fadiscuss

Avast acquired AVG back in 2016. So yes, all that data ends up in the same company.

doh

8 anni fadiscuss

Didn't know it's 2. Pricing may be steeper in comparison, but it provides more than just a hosted PG.

You are absolutely correct that you have to shard your data to take an advantage. It's not equal to vanilla PG. It's an alternative though.

doh

8 anni fadiscuss

Yes. I meant this more as a hosted solution in general.

doh

8 anni fadiscuss

I believe most of those are offered by Citus Cloud [0]. I know it's not the same, but you can spin just a single worker and essentially end with a faster PG than the vanilla PG.

[0] https://www.citusdata.com/product/cloud

doh

8 anni fadiscuss

Sorry to hijack this post, but if anyone is interested in content recognition, we built one of the largest reverse search engines for video/audio [0] and are always looking for skilled engineers to do more.

I think our scale makes things very interesting. We currently index over 20 hours of video every second and to date we've indexed over 11.5B videos.

Feel free to reach out to me directly at [email protected].

[0] https://softwareengineeringdaily.com/2018/06/22/video-search...

comments