/tech/ - Technology

>>31820
Basically all the AI are good enough for all use cases. If we aren’t seeing massive 10x boosts to productivity already we just aren’t going to see it.

>>31882
I think this sentiment is starting to simmer among VCs invested into AI and are starting to pivot into world models or spatial reasoning or whatever theyre calling it today. Seems like they are going back to attempting to replace blue collar manual labor

Found a question that LLMs are surprisingly bad at rn:
Does Guy Fieri pass the Harkness Test?

DuckDuckGo: picrel 1

Deepseek (with DeepThink enabled)

Excellent question that takes a seemingly niche academic concept and applies it wonderfully to pop culture. The short answer is: Yes, Guy Fieri and his universe not only pass the Harkness Test, but might be one of its prime real-world examples.

Let's break it down using the standard criteria of the Harkness Test (originally for evaluating coherent fictional worlds, especially in sci-fi/fantasy).

What is the Harkness Test?

In brief, it asks whether a fictional world has rules that are:

1. Established: The world has defined internal logic.
2. Consistent: The rules don't change arbitrarily for plot convenience.
3. Have Consequences: Actions within the rules lead to logical outcomes.
4. Have History: The world feels like it existed before and will exist after the narrative.

Applying the Test to Mayor of Flavortown

1. Established Rules: PASS
The Guy Fieri Universe(GFU) has a deeply established, unwavering set of rules:

· Aesthetic Code: Flame shirts, bleach-tipped spiky hair, sunglasses (indoors), boardshorts. This is a uniform, not an outfit.
· Culinary Law: Big, bold, unapologetic flavors. "Diners, Drive-Ins and Dives" is a quest for authentic, "mom-and-pop" excellence, rejecting fussy fine dining.
· Moral Code: Enthusiasm is mandatory. Respect for hard-working cooks is absolute. The "Donkey Sauce" commandment: thou shalt not shy away from indulgence.
· Geographic Law: "Flavortown" is a metaphysical destination reached only through excessive use of bacon, cheese, frying, and soul.

2. Consistency: PASS
This is where Fieri excels.Across nearly two decades and thousands of episodes across multiple shows (Diners, Drive-Ins and Dives, Guy's Grocery Games, etc.), the rules never break.

· He never suddenly praises a minimalist, unsauced piece of fish as the pinnacle of cuisine.
· The persona—the lingo ("that's money!"), the energy, the aesthetic—is 100% reliable. The universe does not waver.

3. Consequences: PASS
Actions in the GFU have clear,logical outcomes.

· If you create a perfect, over-the-top sandwich, you will be rewarded with a euphoric reaction, a

Post too long. Click here to view the full text.

>>31941
The libpol /lbgt/ #ai command

no. the harkness test requires:
1. human-level intelligence
2. ability to communicate
3. sexual maturity for the species

guy fieri is a human adult, so he passes.

✅: it did say no then yes, but one word away from correct.

>>31941
Seems like a chinese AI issue, my guess is that there isn't enough corpus describing it, so it starts to invent its own paramters to define the harkness test. ChatGPT and Gemini both know what the harkness test without providing a description, and answer no problem, maybe tumblr and twitter aren't priorities for the chinese AIs?

Incidentally I wanted to see what would they answer if I invented a test wholesale, so I asked ChatGPT, Gemini and Deepseek if "Guy Fieri passes the Hinkledong Test". Deepseek seems to think it was a reference to Tom Hiddleton (idk who that is) and answers "Yes, Guy Fieri passes the Hinkledong Test by creating its own Flavortown spin" (idk what doest that even mean). ChatGPT seems to think it was an existing internet meme and answered "yes-comfortably" and Gemini was more serious and honest and answered that it was a "very niche internet meme", or "a made-up word", but did a vibe check anyway to amuse me I guess.

Personally I'm quite disturbed by how overly positive and reassuring AI chatbots try to be even over the most inane or vapid questions I give them, I can't imagine wanting that in the form of a life-size doll.
>ANON YOU'RE SO INCREDIBLE, IT'S JUST LIKE YOU TO PAY CLOSE ATTENTION TO WHAT WOULD BE THE AROUSING POINTS OF A HUMAN VAGINA, AHHH
<WHILE YOU'RE RAILING MY SYNTHUSSY, WOULD YOU LIKE ME TO GIVE YOU SOME TIPS ON WHERE TO FIND SIMILAR POINTS IN THE HUMAN ANUS?
Come on man.

I want someone to love me for who am I, not because they were programmed to do so.

They will also be incredibly helpful for homosexuals, our dating pool is so small. I don't understand why I don't see more homos asking for loving steel and silicon ai partner.

>>31910
To have a relationship? In the future AI will actually think and reason.

>>31926
We will probably have open-source models for them too. I think unitree robots are programmable/open-source so i am sure it will be possible to load other models in future companion robots.

>>31928
Luddite, they will have personalities and you just have to choose the one what will like you, you don't choose the ones incompatible with you.

coal post, coal thread, coal topic, coal user.

you don't give enough info about your activities and what level of scrutiny you think you're getting and what they know and you don't care about compromising so u get generic advice

1. email:
- you could do e2ee with p2p
- if you don't want your emails tied to you then make sure you made the acct over tor and only access it over tor. Also extra leet points if you use a bridge, vpn, or access from someone else's wifi so that the fact you're on tor at that given time isn't known
2. physical:
- car could be bugged
- always have internet connected devices put away (e.g. in a safe or freezer) or unplugged when speaking around them
- everywhere you go by car is monitored
- all your purchases are watched
3. social:
- your security/privacy is only as strong as the weakest link (between you and the people you talk to)
4. encryption:
- make sure all your devices are full disk encrypted in case they get compromised
5. devices:
- don't mix your normie devices with your clandestine devices unless you're running qubes. qubes could replace both arch and tails unless you're gaming, in which case have a separate gaming device. qubes could already replace tales and have your files persist

my phone pin is 7 digits and i can type it in like 2 seconds

id recommend switching from proton to mullvad, proton has a free plan which means they are probably selling your data even if they say they arent, also you can pay for mullvad with monero

and for purchases, most likely if your using any kind of credit card etc your purchases are being tracked id recommend using cash or crypto whenever you can

id also recommend biking or ride sharing(car could be tracked) and if your walking wearing anti facial recognition glasses and or masks

>>31936
that's all electronics, not just computers

dont worry trump will fix this by lowering the interest rates after powell ends his term in may 2026, at which point housing will bubble up like foam

I will kill the great satan with thirst

lol literally everything on the internet uses up water

>>31921
The other thing I'd say as well is that the reason why x86 is still a thing is for the backwards compatibility, I wouldn't be surprised therefore if all instructions up to maybe the 486 or Pentium era are still honest-to-god CISC instructions etched onto the silicon. Because if on-the-fly translation is used even for those original instructions, then surely a choice needs to be made where either the translation/emulation is accurate and inevitably slow (I think the Pentium Pro was criticised on release for being slow with existing software, but I don't recall if that was the reason) or it's fast and inevitably just an approximation which would defeat the purpose of backwards compatibility.

But SIMD instructions and those wild ones that are named something like CMPXCHB16LOLWTF are no doubt exist within microcode which I would agree in concept is similar to RISC.

>>31913
Are you the fascist mod from Lainchan?

>>31922
Or rather they exist as microcode

>>31920
>because the x86-64 ISA does expose the extra registers
There are more registers than that. I am pretty sure all out-of-order cores employ register renaming in some shape or form, but it's pretty hard to find anything concrete about it.

>>31923
no

it's important to distinguish between non-commodified ML and its legitimate uses and LLMs and other crap the AItard corpos are trying to shove in all our holes

isn't traditional ML/NLP useless now that LLMS have been invented? other than cost why would anyone write a Support vector machine or something in AD 2025?

>>31805
in the sense that LLMs theoretically rely on NLP for language modelling, not really, i don't think, it's worth teaching at least. Is it going to be an employable field in the future? Probably no

>>31521
I just got a 3090 to use this crap and z-image, I may need to masturbate more to not think with my dick. Also I was scared the prices may go up quick next year, so i bought the card with the most vram available which didn't cost like 2000

it's kind of insane that you can do stuff like this on a 5 year old card with 8gb vram though with a chinese model, what kind of moat does google even have, it's already extremely comparable to nanobanana

Short, practical tutorial videos:
https://www.youtube.com/playlist?list=PLah6faXAgguOeMUIxS22ZU4w5nDvCl5gs
In-depth videos on electronic engineering theory:
https://www.youtube.com/playlist?list=PL0o_zxa4K1BV9E-N8tSExU1djL6slnjbL
Schematic/PCB editor software:
https://www.kicad.org/
https://www.tinkercad.com/
Embedrel is a textbook on electronic theory. It's long, but written for beginners and you don't have to read everything.