Christopher Cook

@edandersen How's the audio input latency

So the Macbook Pro M5 Pro running Windows on ARM in a Parallels virtual machine (32GB / 8 Cores) is FASTER than a Surface Pro 11 Snapdragon X (12 cores) in single core and just as fast multicore. Macbook Pro is basically the best Windows ARM machine available, AND in a VM

@0xSero Skill issue are you even using insiders bro

I had to open this abomination today, holy fuck. How can anyone use this?????

@James_M_South Maybe! There's no way they tested any of it before they released it but it looked pretty convincing on the surface. More of a wake up call for me I guess.

@webprofusion First edge case will trip them up. They’ve probably left a database exposed somewhere also

Somebody just published a system that competes with one of our (complex) commercial products on a lot of the features. It's not the same, but it achieves many of the same objectives. They started building theirs 5 days ago. Ours started about 3 yrs ago. I am irked.

@0xSero Eh, why are your GPUs doing anything? You're using GPT5.4 remotely.

It's been running for 6 hours on my first prompt. I have 20+ steering prompts, 25% of them highly detailed. My GPUs have been at 100% utilisation this entire time, my room is 28c boiling hot. I can't describe how crazy it is that this is possible, I am a solo knucklehead imagine what a lab with 1000s of GPUs, and lab equipment can do. Imagine what a nationstate can do. Holy smokes, we are not ready for what is on the horizon.

@James_M_South Lol, I missed that.

@webprofusion That's literally what I used. There's new profiling features

If you value your sanity do not ask Copilot in Visual Studio to profile anything. It's absolutely useless.

@James_M_South We probably just need a right click > profile option in Test Explorer.

@webprofusion I regularly use unit test profiling in Rider; (it’s such a great feature) so I thought I’d give this a go. je regrette

@karpathy @swyx Somebody wrote all over it, I'll give you $50 for it.

Thank you Jensen and NVIDIA! She’s a real beauty! I was told I’d be getting a secret gift, with a hint that it requires 20 amps. (So I knew it had to be good). She’ll make for a beautiful, spacious home for my Dobby the House Elf claw, among lots of other tinkering, thank you!!

@PatrickHeizer I'll wait for the movie lol. Surely by now we can just point the computer at a book and say make the movie.

I am pleading with the forces of the universe to convince Villeneuve to tackle yet another "unfilmable" sci-fi series.

@headinthebox It's tricky phrasing, software by definition doesn't last long and has very little value by itself. However if I use a piece of software to decode ancient texts and publish them, that would have centuries of lasting value. So yeah, don't know what they mean.

What do the doubters see that we don't? Or, do we see something they don't. IMHO, the biggest mystery right now in our field.

@molecularmusing @s4schoener @Danukeru Cool, I have no clue about ps5 Dev. So the regular API is documented or not? Is it just a bunch of method definitions and you have to work out what they do?

@webprofusion @s4schoener @Danukeru No, doesn't use anything internal, only regular API calls.

I find this extremely worrying, with many of people I respect saying things like "I no longer write code" or "let LLMs do it". Why did you start programming? Was it never the journey for you, but only the goal? I genuinely want to understand this, I seem to be the odd one out.

@spaceMonster @James_M_South Ironically this probably counts a slander, might really have to hire some lawyers.

@James_M_South carefull SixLabors licensing lawyers spring on any mention of imagesharp

You know you're doing ok when Chrome has a bug you don't. See that slight clipping?

@elonmusk If it doesn't show up in the VS Code model chooser, it doesn't exist.

What are your initial impressions of Grok 4.20? Major upgrades are still landing every week.

@molecularmusing @s4schoener @Danukeru Off topic but I assume this means discovering and using internal system API calls, surely Sony can update their firmware and break those, since they weren't published or part of an SDK?

@s4schoener @webprofusion @Danukeru I'd be *very* imterested in the outcome. None of the usual known hooking techniques apply, there are no symbols, not even ELFs, the docs and SDKs won't mention any of the obvious things - by reading what's there you wouldn't think it's even possible.

@clare_liguori Can't you point the agent at the changes and say create a set of feature focussed PRs for human review based on these changes. Like literally paste that into chat?

When are we going to get serious tools for stacked pull requests? I generate a TON of code agentically, but then I want to split up the changes into bite-size chunks that can be reviewed by humans across multiple PEs (yes, I still think humans should look at code). My agent splits up the changes into multiple commits, but then what?

An Enterprise customer got in touch asking where the ProgramData folder is on Windows. It's tempting to just answer the question but it's a bit like someone asking "Where do we keep the footguns?". If they don't know, then should they be the one doing it? Is it rude to suggest otherwise?

But could a new dev do it and if not what knowledge is required before they can. Agents are unlikely to invent something never before conceived, but I think we're attributing more to human devs than they typically deserve. People building with agents can be the source of the novel approach, if they are clever inventors already. The agent can build what you tell it, which is what most people are doing. They are not asking for invention, they are asking for implementation.

@webprofusion @Danukeru I can't tell you what's there in the docs and what not (NDAs...), but let's just say nothing in any docs or SDK functions will talk about the things you would need to do to achieve this. Not a single word. Without intrinsic knowledge, the agent just *cannot* infer this.

If you point the agent at your codebase and sdk docs, start by asking it to summarise the purpose and architecture. Get it to write a document for each feature area. See if you agree with what it says, change what you disagree with. You now have a knowledge base relevant to your code. Now ask it to plan a specific small feature or change, see if you agree with the plan, correct it as you would any dev that has just been helicoptered in. Once you have a feature plan, say Start Implementation. If it ultimately gets it wrong, the plan was wrong. Then do the same with something relatively complex, or ask it review code you currently find complex and look for issues. If you don't want to use your real codebase, start a side project and use that, that's easier because you're not "protecting" that code in the same way.

@molecularmusing @Danukeru But you pointed it at the docs first yes? I've pointed the agent at a new RFC and gotten a working implementation of a protocol. You could argue that's new code that's not necessarily been done before, but the ideas were not new because they were in the RFC.