rumblefrog

Will weights programming be the norm for creating new models? To what degree do we still need a full training run vs modifying the weights directly?

I'm worried how we can discern a model's performance from benchmark numbers, would it make it more meaningless as a model can directly improve a certain benchmark via weight programming?

I wonder if the same technique can be applied to model abliteration? It would easily make any open weight models uncensored.

[dupe] https://news.ycombinator.com/item?id=50258521

Jujutsu - https://github.com/jj-vcs/jj

For me personally it's open source I was looking for. But I totally understand if it's not aligned with the business value.

It helps me guard against things like "Hey, this company decided to remove this feature that I use a lot (cough cough, watch together on Plex), or put something behind a paid wall where it was previously free".

With open source, I can at least fork/pin it from some version and merge things from upstream that I want, which is the reason why I was looking at Jellyfin.

Do you intend this to be a freemium alternative akin to Plex, or plan to open source it and more of an alternative to Jellyfin?

This is my first time hearing about Pixel Brite, what's the history behind it?

In the same boat, I have a single seat Google workspace account, and can't figure out how to add Argon to my workspace. Asking Gemini seems to make it hallucinate pages that aren't there.

What an interesting timeline, I bought an Seeed reTerminal E1001 just few weeks ago and loaded it with ESP Home for my daily morning brief, and here Meta is advertising support for this niche hacker product.

Location: New York, USA Remote: Yes, or Hybrid

Willing to relocate: No

Technologies: Go, Rust, TypeScript, gRPC/Protobuf, Kubernetes, Nomad, Docker, Terraform, Consul, AWS, Akamai, Fastly, Cloudflare, MySQL, Redis, MongoDB, InfluxDB, Grafana, Datadog, Linux

Résumé/CV: https://zhenyangli.me/cv.pdf

Email: contact [at] zhenyangli [dot] me

---

Primarily looking for mid-size+ companies at the moment.

Senior software engineer at Disney on the DTC media platform behind ESPN, Disney+, and Hulu. I led the migration off a legacy configuration system ahead of the ESPN D+ Flagship launch, designed intent-based CDN policy controls with Z3 formal verification, and built the platform that optimizes multi-CDN traffic mix. I also drove edge transit routing changes for projected multimillion-dollar egress savings and standardized live events authorization across Akamai, Fastly, CloudFront, and Disney Edge.

I'm glad they specifically called out the prose issue, I was always pinned to Fable 5.1 because I wanted to avoid the unreadableness of other Anthropic models.

These would be great for many places that have storage restrictions on e-bikes, but not normal bikes. So you can easily detach and bring with you before storing it.

Anything more daily human friendly/consumable?

Building an unified admin platform[0] for Source game servers & the alikes. I grew up playing Source games and ran a few, but there were always gaps between all the various OSS plugins you install and were quite fragmented, and never designed to work well with each other.

[0] https://rookhelm.com/

Little hostile with the refutal

The automatically update of my compromised passwords on websites is very impressive, and I wonder how it's achieved.

Guess I was right to be suspicious: https://news.ycombinator.com/item?id=48213860

Location: New York, USA

Remote: Yes

Willing to relocate: No

Technologies: Go, Rust, TypeScript, gRPC/Protobuf, Kubernetes, Nomad, Docker, Terraform, Consul, AWS, Akamai, Fastly, Cloudflare, MySQL, Redis, MongoDB, InfluxDB, Grafana, Datadog, Linux

Résumé/CV: https://zhenyangli.me/cv.pdf

Email: contact [at] zhenyangli [dot] me

---

Senior software engineer with 5+ years of experience building large-scale distributed systems, CDN/edge infrastructure, and developer-facing platforms. Currently at Disney working on the DTC media platform powering ESPN, Disney+, and Hulu - designing intent-based CDN policy controls (with Z3 formal verification), optimizing multi-CDN traffic mix, and integrating with Akamai, Fastly, and CloudFront for live events. Previously a founding engineer at a YC-backed messaging startup (Texts) and ran my own infrastructure-heavy SaaS for several years. Comfortable across the stack but happiest deep in backend/infra work - distributed systems, networking, performance, and tooling that makes other engineers faster. Currently pursuing an M.S. in CS at Georgia Tech (Computing Systems, 4.0).

Wonder if we reached a plateau with the model improvements?

Really appreciate the ability to select effort level again.

Looks interesting, what's their revenue model? Or how do we know it won't be abandoned in the near future?

Location: Raleigh, North Carolina, USA

Remote: Yes

Willing to relocate: Yes - New York

Technologies: Go, Rust, TypeScript, gRPC/Protobuf, Kubernetes, Nomad, Docker, Terraform, Consul, AWS, Akamai, Fastly, Cloudflare, MySQL, Redis, MongoDB, InfluxDB, Grafana, Datadog, Linux

Résumé/CV: https://zhenyangli.me/cv.pdf

Email: contact [at] zhenyangli [dot] me

---

Senior software engineer with 5+ years of experience building large-scale distributed systems, CDN/edge infrastructure, and developer-facing platforms. Currently at Disney working on the DTC media platform powering ESPN, Disney+, and Hulu - designing intent-based CDN policy controls (with Z3 formal verification), optimizing multi-CDN traffic mix, and integrating with Akamai, Fastly, and CloudFront for live events. Previously a founding engineer at a YC-backed messaging startup (Texts) and ran my own infrastructure-heavy SaaS for several years. Comfortable across the stack but happiest deep in backend/infra work - distributed systems, networking, performance, and tooling that makes other engineers faster. Currently pursuing an M.S. in CS at Georgia Tech (Computing Systems, 4.0).

You can also buy very cheap laptop prior, but the hardware wasn't that good... Until MacBook Neo came along.

That number is the total bandwidth for the O2O link.

Easier said than done, vendor lock-in is costly to move from.

I feel like that job would fall on them :P

I tried to deploy a small cluster in the US VA region, but the cluster status kept flipping between Failed and Creating with no clear way of troubleshooting it: 7ad975fb-3c8e-47a9-b03d-9e6bec81f0db

I wonder how long before Hetzner adds something like managed Kubernetes to their native product line. They already have S3 compatible object storages, load balancers and more.