zxspectrum1982

Why does every lab producing an LLM want to have their own harness? Lock-in? Is there any other advantage for the lab?

I'm using Cursor (it's the only way my company allows us to use Grok) and OpenChamber (when using GPT, Muse Spark and others) and I'm happy. If I were to use DeepSeek, I'd use it through OpenChamber too.

* OpenChamber is a GUI for OpenCode.

IME Composer 2.5 is too dumb for any serious coding. Grok 4.6 is twice as expensive (but still much cheaper than Claude Opus) and it does a much much better job.

I use Cursor at work and love it. It's a great IDE, the best IMHO. But it becomes really expensive.

Since the introduction of the Cursor Token Rate, my company tells us to use Claude and Codex directly with Anthropic (or Vertex AI) and OpenAI, respectively.

https://cursor.com/docs/models-and-pricing#cursor-token-rate

I'm now using Grok 4.6 most of the time. Sonnet some times. But that's it. Cursor-native models 99% of the time. I'm pretty sure Anthropic will follow OpenAI soon in pulling their models out of Cursor because it's now a true competitor.

What FPGA are you using? Why?

What exactly did you implement? A full LLM? A subset of it, which collaborates with something running on CPU or GPU? Which LLM? Why?

What language did you use to implement your thing: VHDL, Verilog, Vitis, something else? Why?

I can think of at least 10 blog posts that I'd write before I write a single line of code. Publish early, publish soon ;-)

I'm expecting the Taalas MSIC version to cost a fraction of that. Then probably have some kind of cheap subscription to Anthropic for updates (yes, Taalas chips can receive a certain kind of updates: they have a small SRAM).

1. How come you didn't make your implementation public? You could be a millionaire now. 2. Especially if AMD has the technology to do what Taalas does, it makes a ton of sense for AMD to acquire Taalas: remove them from the market. Make sure nobody else (Intel, Huawei, Alibaba, NVIDIA, etc) acquires them. It could have been a great acquisition for a rebirth of BlackBerry btw.

I'd gladly pay for a Claude Opus 4.6 Thinking High in silicon and use it for 1-2 years. It's good enough for many coding tasks.

Well, then it provides so little they could have not done anything at all as well. Useful stacked PRs would be something that uses AI to split my large branch/PR into smaller PRs for upstream to review. Claude does that for me.

I don't like this.

The case where I need stacked PRs is when I have a ton of changes and I want to upstream them. I have so many changes that I have probably written code in this order: 1. feature1 work 2. feature2 work 3. architecture rework 4. docs 5. feature3 work 6. optimization 7. docs 8. feature4 work 9. security fixes 10. optimization 11. docs 12. last pass security fixes

By the time I want to upstream, I probably want to reorder my commits and generate on PR per theme (arch, feature1 + docs + optimization, feature2+docs + optimization, etc) before I submit a bunch of PRs.

GitHub stacked PRs solve none of my problems. Stacked PRs doesn't take care of the reordering of commits, it doesn't take care of rebasing changes, it adds very little on top of what I was already able to do by saying "this is PR 1 out of 7, this is PR2 out of 7 and build on top of the branch that I used for PR1/7, etc".

Hugely disappointing, bordering useless.

KDE for Enterprise needs single sign-on support, for starters. Gnome has had that for ages.

AI code assistants are lazy junior developers (so you need to keep an eye on everything they do, e.g. make sure they don't forget or ignore to do some parts of the work you tell them to do) with the ability to do research as in-depth as a senior developer. Now make your mind about it. When driving an AI code assistant, you must act like you are a softare development manager managing a team of junior developers.

It only hallucinates if you use it wrong. Give it enough requirements, context, keep an eye on all it does (like it were a junior developer), and you will get awesome results. AI code assistants are a fantastic tool that can make you 10-100x more productive provided that you know how to use them.

Most of the people saying the kind of things you say don't know how to properly use an AI code assistant.

so why did xAI acquire so many GPUs but have no use for them today?

How can SpaceX have so much GPU spare capacity? It doesn't make any sense.

Did Musk blindly order humongous amounts of GPUs years ago before any of us had any sense of the scale this was going to reach?

There's been also a lot of good talent joining xAI lately.

That article is a piece of horseshit. It was indeed the renewables (too much of them) that caused the blackout, as independent and regulatory investigations have revealed.

https://en.wikipedia.org/wiki/2025_Iberian_Peninsula_blackou...

I know how Wine works, thank you. I've used winelib in the past to port Windows software (that I had the source code of) to Linux myself.

My point is the way wine works today is: WinAPI --> winelib --> Linux --> x86

I. e. winelib is reimplementing WineAPI on top of Linux.

What if we could just decompile those Windows functions and recompile them on Linux, or even x86, directly?

The workaround all of these console recompilation projects use is you must have the original game in order to have the binaries (the ones you theoretically decompile and recompile, but actually you take pre-decompiled-recompiled ones by someone else) and assets (graphics, sounds, etc). For Windows applications on Linux, we could do the same: bring your own Windows, then we can decompile and recompile.

I wonder whether this "recompilation" technology could be used to run Windows software on Linux, as an alternative/complementary technique to Wine.

Back in the day (year 2000, until 10-15 years ago) we had Project Odin to dynamically translate Windows software to run on OS/2: https://github.com/netlabsorg/odin32

Don't forget about live speech translation. You could speak one language, the other guy a different language, yet you both heard your language all the time.

Open-source AIOps platform that closes the loop from Kubernetes alert to automated remediation. An LLM investigates incidents live via kubectl, matches a fix from a workflow catalog, and executes it — or escalates with a full RCA. Approval gates, confidence thresholds, and SOC2 audit trails keep humans in control.

Does it work when using Claude inside Cursor? Or is there some something like claude-replay but for Cursor sessions? (I'm out of home, can't try now)

Perfect for POCs, development environments, demos, and simple deployments where a full-scale object storage solution is overkill. It combines Ceph RADOS Gateway (RGW) backed by a standard filesystem and a modern UI into a single, easy-to-deploy container.

S4 provides full S3 API compatibility while requiring minimal resources and configuration.

OpenShift Virtualization on AWS, even as a managed service ("ROSA Virtualization"), has been available for a while on bare metal. Theoretically this enables ROSA Virtualization on EC2, in case you had valid reasons for such a thing.

Easy: ffmpeg discontinues or relicenses some ffmpeg functionality that AWS depends on for those product alines and AWS is screwed. I've seen that happen in other open source projects.

Google is not paying anyone to find bugs. They are running AIs indiscriminately.

IAR C was my first embedded compiler, a long time ago, and it just worked flawlessly.

Why? Why not simply adopt btrfs?

Other than git-friendly, this is how I remember AutoCAD 10 from ancient times.

Same here.

Does Gemini CLI require API access?