ben_w

Mmm. Or improve the OSes themselves.

At this point, they could likely decompile MacOS classic and rework it to give it modern memory protection, virtual memory, preemptive multitasking, and networking.

I assume booting an OS with a mere few megabytes of RAM footprint would only take milliseconds?

The description of vibeOS in that video is half way between what I thought I was replying to and what I suggested.

Doing it in a way that isn't described as "hilarious" is probably also possible in the time since that video came out. Doing it in a way which isn't described as "a security nightmare" or "liable to lose all your data", probably not yet. Not as vibe-coding, at least.

> I want to do MORE with the same amount of folks.

So do I, but I've been in two companies that had to let me go because the investors said they didn't want my role to exist any more. Plus two other roles where the company straight up ran out of money.

It's not always up to us.

> A while back I saw some people saying we would just have LMMs hallucinate an entire operating system every single frame. I guess technically it's now proven to be technically possible, but it would be thousands of times slower while requiring thousands of times more compute (a multi-million x difference from an ordinary OS!).

Gosh. That would be a very silly use of an LLM.

Bad use of any AI, even image/video models which are certainly less wrong than this. Stuff like DLSS show we're getting stuff that could generate a game's graphics frame-by-frame, but even though it's possible to have an entire game environment (not just graphics, everything) done by an AI, such things demo better than they play due to temporal coherency.

That said, having an AI make the code for each thing as you need it, might work somewhat OK eventually. Wouldn't trust it today (security, data integrity, and a tendency to cheat on tests), but eventually it could happen.

> (This is a joke, to be clear.)

Good that you said so, this is annoyingly plausible.

This was true last year, but it would be risky to assume it remains true from one week to the next.

The rate of growth in this sector is "1. move so fast legislators can't keep up; 2. keep going until you hit a limit, then find a way to hack around that limit; 3. GOTO 2".

I have read that comments which appear to have been written by an LLM are auto-filtered. As these appear to mostly consist of quotations of LLM output, I expect this would be why.

I suggest putting them on a blog or something and linking to that, rather than HN comments.

> I wonder how this squares with the recent "NNs are already clean" interpretability finding, which may have also been the impetus for this kind of research. Similar to their work on CNNs, maybe residual-steering is overkill, there might be just a single neuron that "cleanly" corresponds to "evilness" that you can turn up/down?

By my reading of the TL;DR of "NNs are already clean" (given the full thing is a 65-minute read, I've not been through it yet), the result can be rephrased as "it made an ontology different to ours, so we were confused". (Example: we might expect a language model to identify nouns and there to be a noun-neuron, but the neurons actually trigger at punctuation based on the emotional cue from the preceding clause).

Regardless, a one-neuron evil switch is quite clean, possibly (I'm not sure) clean enough for the older processes to find it.

The conundrum: (1) this was only discoverable because of open weight models, and research labs (and governments, if they regulated faster than the glaciers are melting) could be informed by this; (2) yet because it is known, it now means that it's easy for mid-skilled people to download open models, flip the switch to make the model evil, and then setting them lose to do evil things.

Security by obscurity vs. fully automated insecurity. Clichés need not apply.

And all the 420/69 references. And doing "let that sink in" as a live-action bit.

Only if it has a goatee, which is how we know Robert Downey Jr's Iron Man and Benedict Cumberbatch's Doctor Strange were only the heroes of their own narratives, and nobody else's.

Similar, though for me LLMs have recently noticed I'm writing a book.

Weirdly, one of them "thinks" I'm a completely different person (by name) and CCed an old work email address of mine for a company that ceased to exist about 15 years ago. I'd expected them to mix me up with a different person of the same name, but no.

I think some sarcastic responses to Mendeleev at the time actually suggested this.

Why is Tesla a useful example here?

(Also: Same site, but click on earnings instead of market cap).

Both. There's a lot of dumb luck in this world, and me being a software developer at just the right time was luck.

Working smart, living smart, also makes a huge difference. Every 83/month you save off your expenses (rent, more efficient transport, better utility deals) is the same as earning 1k/year more after tax. At lower income levels (but not the lowest because at the very lowest people don't have a choice), investing that over a lifetime turns "just about making ends meet" into "retire 4 years early".