117 karma · joined September 30, 2024
That's an assumption.
And is that what we really care about anymore? Maybe we are ok accepting some "inefficiencies" in the code for the LLM but we get better CPU utilization or cache/RAM usage
Impossible for you, me, and other humans maybe. But not hard to understand for LLMs
> I want to see an LLM reduce a complex multi-million line codebase to 1/10th of its size or less without breaking functionality. That would be impressive!
They can and do, if you ask them to.
If you have tests to validate functionality, it can easily reduce and refactory.
But, why are we caring about LOC again?
I wonder if a refactor is even necessary.
Concepts like KISS, DRY, etc. are largely human-centric. They're meant to make code easier to write, understand, and maintain - for people.
Duplicating code was traditionally "bad" because it increased maintenance costs, risked divergence, and added cognitive overhead for human developers.
But with coding agents, some of those tradeoffs start to change. An agent can find and update duplicate implementations in seconds. It doesn't need to maintain the same mental map of a codebase that a human does.
And if we have good specifications and comprehensive tests, behavioral divergence should be detectable and correctable. In fact, duplication can offer an advantage over shared abstractions: isolation. Changes to one implementation don't risk breaking unrelated consumers of shared code, and each implementation can evolve independently.
Maybe we won't need as many high-level abstractions, or eventually even some of the architectural patterns we consider best practices today.
If the cost of writing and maintaining code approaches zero, perhaps we should optimize less for human readability and more for runtime performance, memory efficiency, isolation, and correctness.
We might even regain some CPU and memory efficiency by eliminating layers of indirection, virtual calls, and abstractions that primarily existed to make code manageable for humans.
Perhaps the optimal architecture for AI-maintained software is fundamentally different from the optimal architecture for human-maintained software.
A lot of our software engineering best practices are based on the economics/ergonomics of human maintenance. What happens when those fundamentally change?
Then you misunderstand how hardware, software, firmware, drivers, and operating systems have worked for the last few decades.
How does your conspiracy theory hold up to that?
GB10 and N1X are both Blackwell cousins. So they've had consumer-level Linux machines for a while now.
This was simply a partnership with Microsoft, so of course development of drivers and firmware focused heavily on Windows. Especially since I'm sure Windows required a LOT of updating to run nicely on ARM as well as the N1X unified memory model.
Even that argument is weak - the DGX Spark is Linux, and the RTX Spark is practically the same exact chip.
Sorry to remind people of HN, but most of the consumer market still runs Windows. This was a joint project to have a consumer-level Windows-on-ARM machine, likely requiring lots of development efforts from both NVIDIA and Microsoft.
So, for these specific laptops, they put 100% of their attention for it to be on Windows for launch to capture as much market mindshare as they can. Local AI, Graphics and creative work, and some gaming.
I doubt Microsoft would have been happy about NVIDIA having Linux on their laptop right at launch.
Market Share by OS for consumer laptops & desktops:
• Windows: 76.65%
• OS X / macOS (Combined): ~15.67% (OS X at 9.86% + macOS at 5.81% due to user-agent reporting quirks)
• Linux: 5.63%
• Chrome OS: 2.04%
> Among nappers, we further defined napping intention based on the question “In general, were these naps planned, or did you fall asleep without meaning to?” Nappers who answered “naps planned” were classified as intentional nappers, and those who answered “fell asleep without meaning to” or “both (some planned/some not)” were classified as unintentional nappers.
Definitely seems like a possible "correlation is not causation".
I would find it hard to believe that sleep is somehow biologically harmful.
So, like your example, excessive or unintentionally napping is instead a sign of an underlying issue or poor health. Possibly the body trying to sleep to repair some damage?
> Zuck: "Dumb fucks"
- 2004
Instead of tech, tobacco, whatever
It's just corporations.
The food and sugar industries have tried to control or pollute scientific narrative. The car industry ("Unsafe at Any Speed"), oil and gas, mining, chemical manufacturers, and on and on.
Why do so many people seem to forget the lessons of the past with some "shiny new" industry, become enamored with it, and think "wow, these are the good guys!".
When billions upon billions, and now in some cases *TRILLIONS* of dollars are at stake, corporations will do whatever they can to protect and grow that money.
And that's not counting AI companies "cannibalizing" their own products, like Sora video generation
And the compute.
Anthropic, Google, OpenAI, Meta, xAI(whatever is it called these days) are sitting on mountains of compute.
Yes, local and open-weight models will continue improving, and they'll be good enough for an increasing number of tasks. But there's a difference between making yesterday's capabilities cheap and pushing the frontier forward.
The most capable models will likely continue to demand enormous compute resources, both for training and inference.
And the pricing is already there - we're charged for token usage, not model access. Draft an email, summarize a paper - basically free. But the "big stuff" will require Frontier-models with Frontier-level of compute.
OpenAI spent something like tens of millions on "solving" a Millenium Prize math model. And there will always be another problem worth throwing massive amounts of compute at, whether it's mathematics, scientific discovery, engineering, etc.
Those are rookie numbers.
I've been doing that for 40 years
Especially as organizations grow in size.
A small startup usually has little-to-no "junk" or "fat" to trim.
But as orgs grow, and since we have imperfect knowledge and communication systems, more an more "junk" or "connectivity" roles emerge.
Microsoft: 223,000 employees
Alphabet / Google: 198,933
Apple: 150,000
Oracle: 141,000
Meta / Facebook: 75,472
I tried to stay way from companies with retail or warehouse workers like Amazon and Walmart (though Apple likely has a lot of retail).
I don't see HOW companies get to that size without some "junk DNA" that keeps it operating.
Maybe in the next few years we'll have AI-powered knowledge bases and communication lanes that smooths things out, but - how do we 'quantify' what "Steve, the guy who has been there for 22 years and works as the company phonebook and knows who to refer you for the thing you need"
Or worse, you removed the workers and left the 'telephone' people.
In my anecdotal experience, MOST companies don't take any time to understand what they are cutting.
They kill "Project X" so the easy answer is that everyone involved with Project X must go. Regardless if they were useful or could be useful elsewhere.
Or they cut by salary, employment length, etc.
How much is that likely to bias the dataset?
So humans do/did still have a role to play
It could have just been a regular black usb stick an I bet nobody would have questioned it.
Instead "why is there something in toilet paper behind the TV?"
If it is so critical, why don't we have tests and benchmarks for it?
Oh right, because "Greg the super-coder" never wrote any of this down and it lives in his head so he can be the arbiter of change.
It was toxic before agentic coding, it's still toxic now.
Document your systems. Add tests and benchmarks for what is important.
We've always known this, we have a dozen or more textbooks written about it. But we always gave it a pass because "Greg the super-coder" was producing more value than he was hurting.
That's no longer the case. Your best coder is now Opus or Astra
But how many changes are actually serious?
I can have an agent mass-produce security bug fixes, zip through our low-priority backlog, etc.
IMO - we need to get out of our own way.
Building is cheap now. Agents are also really good at following checklists and running tests.
So instead of having individual engineers reviewing code, we need to
* Leave human review for the most impactful changes
* Get better at defining the quality bar
* Continually codify human concerns into rules or systems that can be checked against
A lot of this was touched on a few months ago with Spec-driven development becoming a hot topic when agentic coding started.
We should continue that with spec-driven review, testing, etc.
They keep all the input and use it for training data and reinforcement.
If you're being "mean" to it, you're "spoiling" the data you "give" to them - causing them more work to filter it out later.
As with most things in business and life - just follow the money
I don't mind something that requires a modifier key like Shift+click to follow a URL
But so many times now I'll click to get focus and now that's selected some option or opened some pane.
If I wanted that, I'd use the Claude Desktop app!
They are just better at hiding it or configuring Claude.
I have several skills that reformat text to remove AI-speak tells.
Why "even Nvidia"?
They are fully behind using AI for basically everything.
What's next? "Damn, even McDonald's is putting out unhealthy food"
Their skills formats are basically identical, so I setup simlinks from their own skills directories into a shared one so Claude, Codex, Cursor, and anything else that comes out will all read and write to the same shared skills.
It's great having access to the same skills no matter the harness being used
https://www.xbox.com/en-US/accessories/controllers/xbox-adap...
Received high praise if I recall correctly.
Sure, we've all known engineers and product managers that don't consult the target audience, but don't assume that of everyone.
Ok, but that's not a "Silicon Valley" thing.
Go talk to politicians or Home Depot and Lowe's and the home builders conventions.
That's why the post screams of mis-placed anger.
Silicon Valley probably would invent FaceID / NFC / Passkeys / WIFI authentication for doors - but that doesn't mean they'll get installed anywhere.