Hacker Newsnew | past | comments | ask | show | jobs | submit | voidUpdate's commentslogin

So why do humans keep trying to make a better LLM when the LLM is generally intelligent and should be able to make its own smarter LLM?

But that's according to your definition of AGI. According to OP's definition, stochastic word generation is the pinnacle of intelligence, the place where we should "stop waiting for 'our' AGI". Too bad he stopped there for his own.

I understand why people, even technical ones, give in to the hype. I sometimes do too when I look at the agent spinning, talking to itself as if it's an internal monologue and coming up with actions with real side effects on its own.

But when they start blogging or writing about AGI, especially _if_ they are technical, I wonder why don't they go just one, or at most two layers deep and try to understand why this generally intelligent contraption sometimes happily executes `rm -rf /` and then nonchalantly apologizes after the fact. They liken these slips to the same kind of mistakes humans do, but they are fundamentally different.

Human failure modes are predictable from their skill level. Humans also carry emotional baggage and they weigh the cost of the mistakes they bear depending on their privileges.

For an LLM "rm -rf /" is rare and unpredictable, but "I am sorry" after it is almost certain because they were trained on dialogs and they get better score for "I am sorry" than "that wasn't my fault".


Ah, a bum bag!

of course -- it's just another term for a waist-crotch microsatchel.

If the language doesn't matter, why aren't all LLMs outputting assembly code directly for maximum performance?

You could. But thats rarely where the perf hit is. You rarely need more CPU juice than Go provides, and if you do you probably use Rust or C. There is little to gain from dropping down to ASM, as you most likely run on an OS that is more powered than a microkernel, where you would need ultimate low level control.

Where's the irony? They are happy for anyone to be a part of their community, unless you are making other in their community feel unwelcome. It's spelled out in their code of conduct for you. If you're making innuendos that make other people feel unwelcome, you are not welcome

I've seen such communities though. They're basically a light form of struggle sessions.

Whats the expiration date on, say, the windows 95 version of paint? Because it still loads just fine on my current windows computer, does everything I'd expect from a paint program, and gets out of the way when I don't need it anymore

>Whats the expiration date on, say, the windows 95 version of paint? Because it still loads just fine on my current windows computer,

When people say software is "expired", "obsolete", "legacy", "limited shelf life", etc ... they're not talking about whether somebody somewhere in the world has the ability or idiosyncratic desire to continue running that old software.

Instead, it means the software's ability to be practical and usable in relation to the rest of the current computing ecosystem. E.g. In this perspective, Win95 MSPAINT is "expired/obsolete" because it can't open .PNG and .JPEG files. In today's world, .PNG/.JPEG files are much more common than the .BMP files that MSPAINT 95 can open.

I still have an old 16-bit Microsoft C/C++ v 7.0 compiler from 1992. Did it expire?!? Well, not literally. There's no ticking time bomb in it that checks the system clock and refuses to run. Instead, the issue is that it doesn't have the newer C99/C++98/C++11 features and I can't use it to compile today's ffmpeg source code. That's what people mean by "expired/obsolete".


How does one run Windows 95 Paint in 2026? Are there people who earnestly use old school Windows tools?

I don't think OPs comment was meant to be taken literally word for word. Breaking down language this way is hardly a productive way to discuss. OP is saying that most software has an expiry date, is this even controversial? All active software companies are constantly rewriting their stuff and would love to move to a more optimized stack, if cost of porting is not a factor.

I don't know the expiration date of all software, I only know all software has some expiration date. Maybe that version you're using will not be usable in the next 5 years, and it will prove my point. That doesn't mean you cannot use it anymore, it means better options took over.

Id want to be careful of it getting cold and damp if you put a consumer camera inside a fridge. You'd ned one rated for that sort of environment

I would be interested to see a leaderboard comparison between male and female cubers. As far as I know, there should be no real biological advantage either way

Do you know if they also scan for drug residue using those devices? A while back I was travelling through a German airport, and got pulled aside to get the skin on my stomach and the inside of my bag swabbed, and I assume tested in the same kind of machine. I don't know what I did to warrant a further inspection, but I wasn't going to make a fuss. I sometimes wonder what would have happened if they'd pulled aside one of the guys in my group who were definitely using drugs

A piece of software Lockheed made gave its outputs in US Customary units (not following their specification), and NASA expected it in SI units (as their specification expected)

That should have been checked multiple times by NASA before they launched it.

It seems like it should have been checked by Lockheed not NASA since supposedly NASA provided a specification, that specified the units, and paid for the software no?

I just had a quick scan of the incident report (https://llis.nasa.gov/llis_lib/pdf/1009464main1_0641-mr.pdf) and:

> "The output from the SM_FORCES application code as required by a MSOP Project Software Interface Specification (SIS) was to be in metric units of Newtonseconds (N-s)"

(MSOP = Mars Surveyor Operations Program). One of the recommendations was

> "Conduct software audit for specification compliance on all data transferred between JPL and Lockheed Martin Astronautics"

So yes, NASA should have checked the provided software more thoroughly, but also Lockheed should have actually followed the spec they were given. I doubt the SIS is available online to check any harder


No, if NASA chose not to triple check everything before sending something into space then they are as much to blame as the supplier. Blindly assuming something works like it should without testing it makes an ass out of u and me.

Very fitting for the AI age as well. "Yeah I just put the specs of the project you're paying me to do (in NASA's case, probably 100% public funds) but Claude the gimp missed the units because all previous training data use Stones and Yards as units, you should have verified it yourself! I just prompt!"

Should have been checked by both. Lockheed should have checked because of contractual obligations; NASA should have checked because of minimal engineering practices (bugs happen in all your dependencies).

They happen and some are put there on purpose too. In this case it's not a "bug" in your dependencies, and yes NASA since it's using public funding should have been more careful, but ultimately it's Lockheed that was paid to do something no?

I work in aerospace.

Both parties fucked up.

Lockheed's job is to follow the customer's specifications.

NASA's job is to check to make sure what they paid for is what they received.

I do this every single day as a quality inspector here. I don't know why a bunch of highly-degreed engineers can't do a simple job that a person with oonly a GED does without fail.


Actually no, it's Lockheed's that should have had a test-suite and conformance-suite for it, but probably only has a C-Suite. If the program is non-trivial proving its correctness can be very expensive to prove, both money and time wise, and something that is done contrary to the spec is obviously on the one executing the spec and being paid for it, which probably wasn't cheap and probably these expenses add to the "NASA only burns money..." narrative.

"Actually no, it's Lockheed's that should have had a test-suite and conformance-suite for it"

Tell me you don't run AS9100D quality inspections without saying so directly.


Well, it seems like simple test & conformance suites would have been enough to catch this, so you wouldn't need to run AS9100D quality inspections.

Besides it looks like the first AS9100 Standard was released after the incident even happened - perhaps even as a result of this.

So it's totally irrelevant that you both audit software in the space and know the Standard, no pun intended


AS9100 is based off of ISO9001. It doesn't fucking matter, the responsibility is on both parties.

Again, tell me you don't actually handle quality without directly saying so.


yeah Mr. Quality supervisor

That is why the good lord invented acceptance tests.

Would less football be a bad thing? It's the only sport I know where the governing body have to advise the police on how to deal with fan violence, such as "please don't use tear gas on them"

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: