Europe saved humanity
>>101557543>le aiai is a memethey just wanted your data
better version also bump bc fuck hypeniggers from /v/ already
>>101557785>source?ChatGPT 4o is terrible at this.It makes up stuff, and if you ask for a source and provides one and you actually look at the source, more often than not the source doesn't say anything about what ChatGPT is saying.
>>101557826Based, I do that for my writing assignments too
>>101557826its a messwhat else to expect from a statistical modelits 80-90% confident in 80-90% of thingsbut theres 10% of things it has 10% of confidence and it hallucinates things at you based on what youre statistically expecting.with 80-90% certitude too. which also means in 10% of cases it fails miserably
facebook is such a shit company i refuse to believe they could create a competent "AI"
>>101557858It's worse because with a statistical model you could measure the confidence.In an LLM you just get the output you get, that's it. There is no confidence metric.
>>101557858(cont.)btwthats a very interesting concept to explore methinksbecause the act of reinforcing said learning as in validating correct labelsis also part of the data because the preferences of the controller will impact the correlations made by the machinein this casechat gpt not only built from the stats of its part of the conversation,but the prompter's too it has a stat model of my expectations, "shadow data" included in its model.it has been trained on, added through the act of having a bias when validating outputin a way, corectness could be considered as a bias in itself, also introduced into the modelwhich means you should be able to derive the concept of corectness when comparing the models of different controllersits a self adjusting stat modelhow bout someone actually used it as such to probe reality itself?
>>101558122it depends on the architectureim not exactly an ml guru but i read something to of sortdidnt implement it thoi didnt solidify the knowledgei know its a thing thoand cv is the obvious place to lookany classification problem, really
>>101558122>>101558197cont.theres an extremely powerful cognitive tool, the empirical ruleit allows you to derive statistically correct assumptions from partial samplesin absolute you should be able to put a percentage of certitude to everything a stat model outputs.unless its optimized through the destruction of a part of its data.as in:you got a, lets say its 3and b, lets say its 4.the phenomenon, whatever it is, is representable by an addition.so with a =3 and b =4, all you know is c = a+ b.and that c = 7.you will not know with certitude what the a and b werebc the addition is a destructive operationit destroys data.obv extrapolated into maths, or programming as in data thats just not stored.what you see is
>>101557543I love how they tried to frame it as "Europe is missing out". I would pay extra for my apps not to use AI.
>>101558296>what you see is(formatting error)
>>101558315>I would pay extra for my apps not to use AI.Sorry, best they can do is use AI without telling you.
https://youtu.be/aVvkUuskmLY
>>101558386>a beasthmmmm... satan coded?
>>101558374Isn’t it the other way around? ”AI” being just underpaid workers in some unfortunate shithole
if you've made the mistake of having an instagram account you can go and tell them to not use your data for meta ai
>>101558496Nah ChatGPT is too fast to be someone behind a keyboard.
>>101559148True, but apparently some kenyans do the censoring of questionable output