[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: Untitled.jpg (359 KB, 1632x1740)
359 KB JPG
It isn't slowing down
>>
>>108784701
Sure, if you use log scales.
>>
>>108784701
pretty sure i could 80% successfully debug a small python library in 1.5 hours. whats so special about AI?
and i could implement a simple web server 100% of the time in about 10 minutes with probably one? two? imports. EZ.
But i am just guessing (i'm using a logical regression prediction) because i havent ever done those things
>>
>>108784730
the y axis is the length of time the human-subjects took. the llm isn't actually taking that long to do the same task.
it's relatively trivial now to make llms the work on a task for dozens of hours.
>>
>>108784701
Yeah. Maybe in two weeks we will finally have a usecase for it.
>>
>>108784750
i dont understand
you are saying it takes humans 30 minutes to make a web server
but AI can do it in a fraction of that time
and it can run for 16 hours before exploding
so why hasn't it made windows 12 yet or cured cancer or done something impressive.
what the hell is it playing at
>>
>>108784764
>is in /g/
>sees zero use cases for AI
How would you feel if you did not have breakfast this morning?
>>
>>108784777
>you are saying it takes humans 30 minutes to make a web server
>but AI can do it in a fraction of that time
yes
>and it can run for 16 hours before exploding
current with codex and /goal mode we've seen consumers doing 50+ hour runs

>so why hasn't it made windows 12 yet
probably takes millions of man-hours to do and the models aren't capable enough to bother running the experiment yet

>or cured cancer
same as above + also constrained by lack of physical experiment harnesses (being worked on)

>what the hell is it playing at
believe it or not, METR is actually a safety eval.
>METR’s mission is to develop scientific methods to assess catastrophic risks stemming from AI systems’ autonomous capabilities and enable good decision-making about their development.
>>
>>108784792
>the models aren't capable enough to bother
Is "capability to bother" a technical term
>>
>>108784777
because all those "metrics" and "benchmarks" are just elaborate ways AIbros have found to convince investors into giving them more money to throw in the furnace. Hence why you see things like "MYTHOS HAS FOUND 5000 SERIOUS EXPLOITS IN FIREFOX IN 15 HOURS (15 hours of 10000 AI instances running in parallel in our multitrillion dollar datacenter and only after you modify Firefox's source code to disable all security features that would have made those exploits impractical and only after a security researcher has read the source code and told the AI models exactly where to look and what kind of exploits they're more likely to find + some old ones from 10 years ago)"
>>
still waiting for software to get better
any day now I'm sure
so far it's still getting worse tho
>>
>>108784701
VibeGODS winning once again. Codetrans seething rn
>>
>>108784730
>bro what's so special about computer doing what was deemed impossible less than 10 years ago lmao
>>
>>108786709
>deemed impossible
nobody said this
>>
>>108786744
>umm acktschully we already predicted AGI like in 1950s okay chuddy
reverse prognosis at it's finest
>>
Indians worshipping LLMs is like the tribespeople worshipping the Coke bottle in that old movie, The Gods Must Be Crazy.
>>
>>108786839
The only difference is that ai is really useful and it’s replacing codetrans
>>
>>108784701
I hope it doesn't, because we aren't going to see explosive growth until it gets to the point where it can successfully do something which would take a person years, not just a few hours.
>>
this
>>108786240
seems more plausible that this
>>108784792

>>108786709
hey you know that one thing you kind of enjoyed doing that we have been steadily ruining and sucking all joy and fun out of for the past 20 years by inexplicably encouraging the worst people in the world to participate, well good news! we managed to write a computer software that makes EVEN WORSE code! and we are going to shoe horn it into absolutely everything and convince the most stupid people in the world (managers) that if you don't use it, you are a bad person.

so now that all distractions are removed, you have no option but to spend even more time of your pointless fucking shit existence concentrating on how this amazing progress we have made as a society has somehow made everyone completely miserable.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.