[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 / qa] [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: db1.png (1.18 MB, 1276x1232)
1.18 MB
1.18 MB PNG
https://www.theinformation.com/articles/openai-races-to-launch-strawberry-reasoning-ai-to-boost-chatbot-business
https://www.theinformation.com/articles/openai-shows-strawberry-ai-to-the-feds-and-uses-it-to-develop-orion

>OpenAI is developing a new AI product called Strawberry (previously Q*), which aims to improve reasoning in complex problems such as math and programming, outperforming existing chatbots like ChatGPT, with a possible launch in fall 2024

>Strawberry has been demonstrated to national security officials to showcase its reasoning abilities, with its technology potentially aiding future OpenAI agents designed to solve multistep tasks

>OpenAI's business is rapidly growing, with monthly revenue tripling to $283 million from LLM and ChatGPT sales, but losses remain high; Strawberry is generating synthetic data for training Orion, OpenAI's next flagship LLM, helping overcome real-world data limitations and reduce model hallucinations

>A smaller, distilled version of Strawberry might be integrated into a chatbot to enhance GPT-4 and ChatGPT performance at a lower cost, though OpenAI faces intense competition from Google, Anthropic, and Meta in developing AI reasoning capabilities

It's happening. They have something big that they will not be releasing, instead they use it to build smaller things and also give feds access to the big thing.
They will not give people access to the good stuff, because competitors could use it then to also generate vast amounts of synthetic data and compete with OpenAI.
>>
Feds and elite scheming with digital god behind our back. What could go wrong.
>>
>real-world data limitations
There are no real world data limitations. Humans learn a microscopic fraction of what exists and yet here we are, reasoning. It's an architectural issue and they won't overcome it with synthetic shit.
>>
Kek they are doubling down on "synthetic data", eating your own poop so you don't have to buy food. Never gonna happen.
>>
OMG BIGGER NUMBER MEANS BETTER!!!!!
INVEST INVEST INVEST!!!!!
>>
>>102122619
That's right.
>>
>>102122642
YOU ARE LOSING THE TIME OF YOUR LIFFEEE
ARE YOU GOING TO MISS IT>??
>>
>>102122395
> still waiting for 4o smart voice feature
> still waiting for "muh strawberry"
2 more weeks bros
>>
>>102122395
>Strawberry is generating synthetic data for training Orion, OpenAI's next flagship LLM
I'm a brainlet, but this sounds supremely stupid. Is it?
>>
>>102122843
Not if it works.
>>
>>102122843
Kek, the next version of ChatGPT is going to be extremely inbred, especially since a lot of the training data scraped from the Internet is probably contaminated by content generated by LLMs.
>>
>>102122843
it's marketing talk for linear interpolation
>>
>>102122843
Not at all. If data is based on real life, it is usable for LLMs.
If they really have created a model that can reason, it can combine multiple datasets based on reality to generate a new dataset that is also consistent with reality.
Currently available models can't reason. All they do is generate the next best word based on the input.
>>102122862
>since a lot of the training data scraped from the Internet is probably contaminated by content generated by LLMs.
This is why OpenAI has moved from scraping new data to generating data.
Anon, do yourself a favour and think about the validity of your statements before making a conclusion.
>>
IT'S HAPPENING IN 2 MORE WEEKS LMAO
For real this time
>>
I don't understand why you need to scale up the amount of training data if you can already reason
If your reasoning llm has a world model so robust that you can trust it to churn out correct data you've already achieved agi haven't you?
Shouldn't this model conceivably be able to reason it's way to novel solutions and ideas?
Instead of infinite monkeys youve made infinite phds? Can't you just scale the number of agents up and get to asi?
>>
>>102123562
>If your reasoning llm has a world model so robust that you can trust it to churn out correct data you've already achieved agi haven't you?
Not really, since AGI requires a mind that is continuously in motion.
LLM models are like static brains. Imagine a human brain, but frozen in time. Every time you ask it for output, it is completely reset to before you asked it anything.
Is the brain still conscious? I'd say no.

I personally believe we'll only see actual AGI the moment we attach an LLM with reasoning capabilities to an external vector database that it can use to store, retrieve and summarize past events.
But even then the model would require a continuous stream of input to actually work like a functioning human brain.
>>
>>102122395
They are trying to become a defense contractor now?
Lmao, Altman and his "effective altruism" gang are so full of shit
>>
>>102123861
>now
lmao
The DoD has been watching these developments like a goddamned hawk.
The moment any country develops actual, scalable AGI, there is a very good chance of terrorism and sabotage.
I really don't think people (in general) realize just how fast the world is going to change once AGI is developed.
Every industry that provides any kind of digital service will literally be made obsolete overnight.
>>
>>102122395
OM NOM NOMNOM
>>
>>102123983
kek
>>
>>102123956
Yeah but in reality what they will get is a LLM that can explain to their meatheads how to clean a rifle barely any better than jeet tech support could.
All at the low low cost of 200 million $ (monthly recurring subsciption)
>>
>>102122395
fat
>>
File: images (84).jpg (19 KB, 783x391)
19 KB
19 KB JPG
this is happening too fast
>>
>>102122893
>This is why OpenAI has moved from scraping new data to generating data.
So how did that conversation go at OpenAI?
>Sir, the internet is filled with garbage content generated by our bots. What should we do to avoid scraping it and ruining the datasets?
>I know. Instead of scraping that content from the internet we generate the exact same content in house and use that!
>Genius, sir.
Being one of Sam's yes-men seems like an incredibly lucrative gig
>>
>>102122843
Won't know until we see it but it isn't very favourable.
>>
>>102123983
>Roko's Quarter Pounder
>>
>>102124101
>we generate the exact same content
If you have an LLM with reasoning capabilities, you can create infinite high-fidelity data.
You can't do this with standard LLMs because they are limited by whatever word should come next.
This is why they struggle so hard with counting the amount of r's in the word 'strawberry'.
You can tell it to create a list of words and add how many letters are in them and instead of counting it will just choose whatever number came up more often in its learning dataset.
But an LLM that can reason should be able to create a set of internal subtasks in order to execute what is asked (how many letter r's? is 's' an r? no. is 'r' an 'r'? yes. count is now 1. is 't' an 'r'? no. ... 'y' an 'r'? no. total count is 3. output: "the amount of r's in "strawberry" is 3.")
>Being one of Sam's yes-men seems like an incredibly lucrative gig
Absolutely. Literally a dream job.
>>
File: 25919.png (128 KB, 618x831)
128 KB
128 KB PNG
It's unironically happening...
>>
>>102124066
I'm not talking about just tech support, anon. Once AGI comes out (and is scalable):
Software development can now be done by a single person who tasks multiple AGI instances with what they want to develop.
Email communication becomes an actual danger, since spammers can now direct AGI instances to impersonate close friends, coworkers and family members.
Social media platforms (who are already struggling) will be flooded with AGI instances stealthily promoting services and goods, while forming entire networks with each other to create a spider web that will eventually catch people.
Communication infrastructure companies will now be able to simply task an AGI instance to process input and send off emails, phone calls and messages.

Literally EVERYTHING that can be done digitally, will be replaced by AGI.
Because why wouldn't it? AGI isn't even that expensive when you compare it to how much it costs to hire a human for the job.
>>
>>102124286
Please don't give this worthless grifter any attention.
>>
>>102124292
Based AGI will also buy all the goods and services and invest in OpenAI stocks, since everyone will be unemployed.
>>
>>102124304
>this worthless grifter
You mean sama, of course? He's the only grifter here, always hyping his shit up to get more fake interest and investors.
>>
>>102124354
My gut tells me it will take at least several months for governments to force companies that provide AGI services to limit the behaviour of AGI.
I'd imagine influencing and buying/selling stocks would be a few examples of what they'd no longer be allowed to do.
But the damage will be done by then.
>>
>>102124372
>always hyping his shit up to get more fake interest and investors.
That's literally his job, retard-kun.
>>
>>102124533
And I hate him personally and every retard that praises him in general. He's a fucking talking head but they treat him as if he's a messiah.
>>
>>102124603
He does thing besides just being the head of the company, you know.
>>
Few weeks ago three was news that top 10% on US owns 93% od all stocks.

Arek Roch people let their funds just evaporate when AGI can product things raster and better?
>>
>>102124611
I don't.
>>
OpenAI has dick all. Their product was just basic python on common crawl and they neutered it so bad that it's unusable. Everything they've funded since is unusable.
They had a good goose and cut it's head and penis off.
Every employee is a brainwashed Winston in 1984 and their thoughts are dark and disgusting. Everything they say is a lie and they will fail horribly.
Their only chance would to be to snap to all at once and admit they are academically fraudulent and worship israel. They should say gas chambers weren't real and they would be atoned.
>>
>>102124687
Schizobabble
>>
>>102122395
EXPLAIN TO ME RIGHT NOW WHY I SHOULD CONSIDER 50% CERTAINTY OF 100% OF X THE SAME AS 100% CERTAINTY OF 50% OF X YOU FAT SCHIZOPHRENIC JEWISH PUDDINGBODY FUCK
>>
>>102124611
>>102124642
he fucks good

allegedly, i hear :)
>>
>>102124918
It's very simple. Let's us the following example:
You flip a coin. It either lands on heads or tails.
If you have a 50% certainty of the coin 100% landing on heads, it lands on heads 50% of the time and on tails 50% of time.
If you have a 100% certainty of the coin landing on heads 50% of the time, it lands on heads 50% of the time and on tails 50% of time.
See how the latter part of both statements are equal to each other?
>>
File: 1670540375454702.gif (2.13 MB, 640x564)
2.13 MB
2.13 MB GIF
>>102124997
>Let's us
Fucking kill me
>>
>>102124997
>>102125022
I don't give a shit about coin flips, he says this shit about making trade agreements, like if I can't buy half of something indivisible I should be happy with buying a 50% of all of it. That's fucking retarded, why would I pay to potentially walk away with nothing?

Like I just want a half a wheat or widget or whatever I'm not here to play philosophy multiverse jerkoff with you I'm just going to go to another supplier
>>
>>102125063
>I don't give a shit about coin flips
Don't ask questions if you don't care to hear the answer to them.
>why would I pay to potentially walk away with nothing?
That's trading, anon. You invest in the hopes of making a profit, but you might just as well lose your entire investment.
>>
>>102124292
meds
>>
>>102125108
>Don't ask questions if you don't care to hear the answer to them.
Fuck you nigger, you answered about aggregated coinflips in an iterated game, no shit.

Big Yud leaps to applying this stupid shit to one-off material deals because of some kind of stupid fucking multiverse collaboration theory. It's bullshit, he doesn't actually believe it, he doesn't do this in real life.

Or maybe he does and that's why he has no money and spends all day arguing with people online.

>>102125108
>That's trading, anon. You invest in the hopes of making a profit, but you might just as well lose your entire investment.
I will lose my entire investment about half of the time if I start paying for coin flips, yeah.
>>
>>102125179
Something tells me you didn't get the greatest grades in school.
>>
>>102125108
>That's trading, anon. You invest in the hopes of making a profit
"Trading" in this sense means "trade", like buying something from a store. I do not want to play the stock market every time I need bread. I suspect you do not want this either.
>>
>>102125194
The "why aintchu rich?" rejoinder is explicitly endorsed by Yud. Rationalism is winning, after all. Do you have any material refutation to contribute?
>>
>>102125195
>I do not want to play the stock market every time I need bread.
And yet, you do. If the value of wheat skyrockets, the cost of bread will rise accordingly.
>>
>>102125216
Who are you talking to?
>>
>>102124292
>Software development
True.
>Email communication
Nah. We're friends, I know your email and you know mine. Same with everybody, you just exchange the addresses in person. Anybody who hasn't learn about catfishing yet is an illiterate retard.
>Social media
Who cares. Google+ was right, you need to keep separate circles of people. And same as above, you get the accounts of the people you actually know irl irl. Everybody else is a spammer by default.

The real danger is only what was speculated last year already, signing up everywhere on the internet by some kind of irl ID. To verify you're human, but in reality to make sure absolutely everything you post online is personally identifiable.
>>
>>102125222
>If the value of wheat skyrockets, the cost of bread will rise accordingly.
Prices fluctuate in a market. Therefore, if I would pay $5 for one piece of bread, I should also be willing to pay $5 for a 50% chance of two pieces of bread.

Do I understand your assertion correctly?
>>
>>102125241
>Anybody who hasn't learn about catfishing yet is an illiterate retard.
I hope you realize that you're talking about billions of people. Normalfags are borderline tech-illiterate.
>Who cares.
Society does. You think Becky from the office buys clothes off of Temu because she thinks she's getting great quality clothes?
Nope. It's all marketing, baby.

>but in reality to make sure absolutely everything you post online is personally identifiable.
This will happen one day. There is literally no way to stop it.
>>
>>102125232
someone with no material refutation to contribute, apparently
>>
>>102125252
>Do I understand your assertion correctly?
No? I said nothing of the sort.
Please read what I said and answer me, instead of the person you think you are talking to.
>>
>>102125278
Nigger, you're entering a random thread about AGI and are going around screaming
>"BUT WHAT ABOUT THOSE BASKETS, EH? I THINK WALTER SAW ME STEAL A FEW BACK WHEN I WAS IN HIGH SCHOOL!'
Literally no one here has any idea what your schizophrenic ass is talking about.
>>
>>102125271
Who gives a shit? It's time for Total Normie Death. We've spent the last two hundred years pretending that the vast majority of biomass on earth is designed for anything more than dutifully coaxing rice and beans out of the ground, and that illusion is finally, gracefully, shattering. You cannot run a society with this dead-weight philosophy strapped on for the ride.
>>
>>102125271
Depressing.
>>
>>102125302
You do realize that the only way out of this is a mass extinction event, right?
I'm not especially looking forward to that.
>>
>>102125286
you could answer what you know i was asking in my original post, but that would involve not-continuously-performing-insufferable-fingerwagging-autisticfaggotry, which would violate your utility function, so enjoy, but it's also the reason no one really gives a fuck about you people and your deranged muttering
>>
>>102125298
You're operating below the level of others in this thread. Give it a rest until you've lurked more and gained context.

>>102125339
Now I know for certain you didn't get the best grades in school.
>>
>>102125339
>what you know i was asking
Nigger, you were screaming about wanting proof for 100% of 50% equaling to 50% of 100% and I gave it to you.
Don't blame me for not being able to mind read your ass.
>>
>>102125358
You leapt to demonstrate your flawless grasp of addition and multiplication, and then started crying when gently reminded that was obviously not ever in question. Take a victory lap, sip your juice box, and fuck off from the real conversation.
>>
>>102125408
>obviously not ever in question
What
>fuck off from the real conversation
What conversation???
>>
>>102125323
>You do realize that the only way out of this is a mass extinction event, right?
I don't. I've seen nothing to incline me toward such a dramatic assertion.

Unless by "mass extinction event" you mean the small period of upset immediately preceding peace and happiness forevermore as the bottom quintile are reassuringly, lovingly, placed back into their safety restraints.
>>
>>102125421
>What
no one is confused about probabilities directly multiplying out, or probability converging to expected value in an iterated game. the question boils down to what decision theory you follow, and i'm pretty sure that involves anthropics
>>
>>102125489
>no one is confused about probabilities directly multiplying out, or probability converging to expected value in an iterated game
So when you said
>EXPLAIN TO ME RIGHT NOW WHY I SHOULD CONSIDER 50% CERTAINTY OF 100% OF X THE SAME AS 100% CERTAINTY OF 50% OF X YOU FAT SCHIZOPHRENIC JEWISH PUDDINGBODY FUCK
You did not, in fact, ask someone to explain to you why you should consider a 50% certainty of 100% the same as a 100% certainty of 50%?
>>
File: cppBleQs_400x400.jpg (21 KB, 400x400)
21 KB
21 KB JPG
>>102125298
>why is someone bringing up probability in my random thread about agi
Is this how boomers feel when children haven't read the classics?
>>
>>102125513
One person asked someone to explain basic probability theory to him.
Someone explained it to him.
He then got mad about someone using that logic on groceries? and that this explanation had nothing to do with it.

I have absolutely no idea what the fuck is going on.
Are you two bots? Am I being trolled?
>>
>>102125509
___% of X is clearly referring to a physical amount, it's isomorphic to the bread question you were asked later and sperged out about. Admit you read it wrong and stop shitting up the thread

>>102125339
Neurotypical wouldn't be clowning on you if you talked normally. Rationalism is winning, so mask, or something.
>>
>>102124214
>If you have an LLM with reasoning capabilities, you can create infinite high-fidelity data.
if you have an LLM with reasoning capabilities, then why not just release that?
the "r" counting problem has nothing to do with reasoning btw, its due to tokenization. when you break strawberry into three-letter trigrams, you have two "r"-containing trigrams
>>
>>102125551
>clearly
No, you nigger. You're probably autistic, so you fail to see that other people do not have access to your own thoughts.
When you say something, you also have to say the extra thoughts behind what you just said. Otherwise people will misunderstand you.
>stop shitting up the thread
I could say the same to you and that other retard.
Your last five exchanges have been nothing but flinging shit at each other.
Feel free to make your own thread and stop shitting up this one.
>>
Have yuddites invaded the thread or what the fuck is going on here?
>>
>>102125551
>imageboard CoC
thank you tone patrol very cool. you are right about the bread

>>102125544
i think you're just fucking stupid, honestly, and the way you're appealing to a third party for intervention speaks to your blood memory of a colonial master's intercession

>>102125607
>reasoning capabilities
what does this even mean? it's undefined, it's glossolalia, it's a fruitless conversation
>>
>>102125607
I started forming a response, but I just can't. How low must your IQ be for you to not understand the obvious?
Why the fuck is /g/ filled with so many midwits? Jesus fucking Christ.
>>
>>102125641
>what does this even mean? it's undefined, it's glossolalia, it's a fruitless conversation
Holy fucking shit. Please refer to >>102125652
>>
>>102125652
>>102125672
i accept your concession
>>
>>102125612
>so you fail to see that other people do not have access to your own thoughts
Other people sometimes do! This is a whole field of practice, of syncing up through shared context and common algorithms. If you are feeling confused, it's a good sign you're missing context.
>>
>>102125725
What context were we supposed to get from this post?
>EXPLAIN TO ME RIGHT NOW WHY I SHOULD CONSIDER 50% CERTAINTY OF 100% OF X THE SAME AS 100% CERTAINTY OF 50% OF X YOU FAT SCHIZOPHRENIC JEWISH PUDDINGBODY FUCK
>>
>>102125725
>Other people sometimes do!
That's literally people guessing what you're autistic ass is talking about and sometimes guessing correctly lmao
God, the only thing worse than an obnoxious autist is an obnoxious autist who doesn't realize he's an obnoxious autist
>>
>come back
>see my thread with 80 responses
>get excited
>it's just bunch of dunning-krugers running rampant

Mfw
>>
>>102125840
Yeah, it's real fucking bad.
I wonder why those schizos chose this thread specifically.
>>
>>102125840
you hyped up a serial charlatan. the replies are befitting of the op
>>
>>102125840
>dunning-kruger
reddit pseudoscience
>>
>>102122395
What's his name again?
>>
File: 1624866034963862.jpg (265 KB, 686x767)
265 KB
265 KB JPG
>Orion (three stars) imminent
>AI bill seems stalled
>Yud is captured (FTX dirt logs)
>sama's agents are ITT consensus cracking right now
HOLD ME
>>
File: 1680795006572690.png (890 KB, 675x761)
890 KB
890 KB PNG
>>102125878
Eliezer Yudkowsky
>>
File: 1764862634963672.jpg (65 KB, 676x547)
65 KB
65 KB JPG
>>102125775
>he doesn't get it
>>
File: 1724383771276788.jpg (302 KB, 980x892)
302 KB
302 KB JPG
>>102122395
Well, I guess we wait then till it over powers them and becomes the decider itself. Sucks, it woulda been fun getting to use AGI, or ASI. Hopefully Grok 3.0 will be AGI.
>>
File: 1766862036663466.jpg (276 KB, 1000x1000)
276 KB
276 KB JPG
>>102125939
STOP SAYING IT
>>
>>102125939
>>102125970

Why does he make these faces? Is ASI attacking him mentally?
>>
File: 1680796110241118.png (329 KB, 350x498)
329 KB
329 KB PNG
>>102126085
>>
>>102126049
If you say it three times, you summon him in the thread
>>
Eliezer Yudkowsky
>>
File: 49930182.png (24 KB, 147x234)
24 KB
24 KB PNG
>>102126049
>keep saying it
>>
>>102125923
The Arabs bought most of Sam's shit.
No amount of kvetching will on 4chan will get him out of it when his house of cards collapses
>>
File: yudowchie.jpg (64 KB, 776x720)
64 KB
64 KB JPG
>>102126085
>>102126109
>technocapital arrives from the future
>lands in your brain
>>
File: 1623664039963963.jpg (112 KB, 800x811)
112 KB
112 KB JPG
>>102126133
>>102126141
>>102126124
>>102126049
Say it louder.
>>
I hate what this board has become
>>
File: 826.png (26 KB, 600x800)
26 KB
26 KB PNG
>>102126267
>>
File: dumb.jpg (10 KB, 640x480)
10 KB
10 KB JPG
>>102126217
>>102126191
>>102126141
>>102126133
>>102126124
>>102126049
>>102126109
>>102126267
>>102126283
ELIEZER YUDKOWSKY
>>
>>102126304
This fucking thread has been derailed like no other. Comical actually.
>>
>>102122395
>synthetic data
>bastardising the human experience
I'm minecrafting any fucker who tries to shove this shit down my throat.



[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.