toshinoriyagi

I am interested in seeing how much these cybersecurity capabilities correlate to general programming. Cybersecurity definitely feels like it would be easier for an agent due to the natural explicit feedback "did I get access or not". While general programming has many less-explicit concerns (is the code readable/maintainable, robust, bug-free, performant, scalable etc).

I don't think this is accurate. Easy Anti-Cheat has supported Linux for a long time, many years before this. I doubt they were years and years out of date, updated, and allowing Linux was turned on by default in the new version. What they likely mean is they ticked the box in this game update to let it run on Linux.

Why would they specifically mention the Steam Deck? EAC doesn't have separate flags for Linux and Steam Deck. It's almost certain they intentionally allowed it to run on Linux. They call out Steam Deck by name because it is a popular handheld that has greatly increased the # of Linux users, not because an EAC update allowed it to work by chance.

I don't think this is true. I've been using NixOS for years and it's awesome. Total game changer for managing my PC and development environments. More adoption will help of course, but I certainly haven't needed it to get tons of value from Nix.

This model is a preview of Qwen's upcoming Qwen4 architecture. It is a 125B-A6B MoE model, meaning it has 125B total parameters with 6B active at a time, but it also has a 51B parameter engram with it. The engram is basically a lookup table for tokens to my understanding. It allows the model to have access to a much larger amount of info if utilized well.

They said the model is intentionally under-trained since it is mainly for R&D purposes of proving the new architecture. Many people are excited for models in this range as they are a step above the common ~27B models, while not requiring exorbitant sums of money to run like much larger models.

Yup, direnv + Nix. Incredible combo for development.

Sure, but a seed oil is not a plant. The oil is extracted from a plant. You can eat all the plants you want, but unless you directly consume the processed extract, you won't be getting much fat. Avocados are a source of healthy fats. However, this still leaves you deficient in EPA. Conversion from ALA is trash, so the only option to my knowledge is seaweed/algae and you would massively over-consume other nutrients trying to hit the DRI for EPA. Your only option is supplementing with algae pills I think. From a quick glance, this is 2-3x as expensive as fish/fish oil.

Tofu, seitan, and tempeh all have protein, but they differ from animal protein in that they are "cut" with lots of carbs and fat. I am on a protein-sparing modified fast (PSMF) for bodybuilding goals right now, and there are no viable non-animal whole sources of protein (so, excluding pea protein isolate etc.) I can have. Even seitan is only ~81% protein by calories, which is too low. I am targeting 90% calories from protein right now.

They are definitely good options, along with lentils (which have a similar problem of having more carbs than protein), for vegetarianism+. But they are not a perfect substitute as you cannot get nearly pure protein from them. They also tend to lack the BCAAs that make up a large portion of muscle.

So, you can fulfill most dietary goals purely vegan, but it is harder and more expensive due to limited options.

Various plants (dark leafy greens/cruciferous vegetables) do make up the foundation of a healthy diet, but do not provide significant protein or fats. Meat is very useful from a dietary perspective. You can make a quality vegan/vegetarian diet, but hitting high protein goals for those who are active can be difficult. The majority of vegetarian+ protein options are also high in carbs.

Plus, meat is tasty. It makes perfect sense people look forward to lab-grown variants that are ethically sourced and have no adverse health effects.

Yes, thank you, that is the correct terminology, will edit my post.

The value proposition for the frontier labs is starting to look really bad. There have been multiple open-weight, Chinese models released recently that perform extremely well, often at much cheaper prices. DeepSeek pricing is insanely low, granted it will go up soon, but with Pro being a frontier level model at $0.435/$0.003625/$0.87 for input/cached/output it's hard to compete. Granted they train on data, but there are other providers that are still much cheaper than frontier labs and don't. Plus these models are open.

GPT 5.6 Luna is the only non-Chinese model that seems to compete at such incredible price/performance. It's concise too, Claude models are wildly verbose. And Google is nowhere to be found in LLMs right now.

Nixpkgs is one of, if not the largest, package managers by # of packages. I think naturally that will result in more abandoned packages than others. What is the % of abandoned packages relative to total I can't say.

Nix tends to have updates almost as quickly as Arch does, I run a lot of bleeding edge software and have no issues, and the breadth of packages is immense. Nix just has a different style. With the AUR, you can adopt an abandoned package more easily. That has pros and cons (the AUR has recently suffered multiple supply-chain attacks due to packages being adopted and infected).

This was a governance structure that began ~1 year ago. Nix is 23 years old, so it won't be dying or going away due to this. But it is sad to see two extremely valuable contributors leave the project due to burn out.

Nix and NixOS are two of the most revolutionary pieces of software in my experience. Hopefully Nix finds a governance structure worth of it.

Deepmind always worked on some of the coolest architectures. Following things like AlphaStar, AlphaGo, and more were extremely exciting and felt like the hacker persona of machine learning. I hope Google can take advantage of this awesome team. They've done incredible work.

They've also announced Qwen3.8-27B being released open-weight next week. Qwen3.6-27B is widely regarded as one of the best local models, especially since nothing else comes close to it, that isn't benchmaxxed, without being significantly larger. If 3.8 truly improves upon it that would be awesome.

No, but they can transfer them, which is what the comment you replied to was worried about. My partner used to be an elementary school teacher and frequently complained about the school she worked at. The district transferred a large percentage of students with IEPs (individualized education program, a plan for special care/resources for students with disabilities, often related to poor behavior) from other schools in the district to hers.

Her school did not have adequate resources to handle these students, so they always had multiple students with severe behavioral issues that should have been in a dedicated classroom with a special education trained teacher, but were just in regular teachers' classes. Naturally, the teachers were burnt out from working with too many challenging kids they were not trained to take care of and the other students had worse learning outcomes.