Last calendar month , Reutersreportedthat ChatGPT God Almighty OpenAI is work on a projectcodenamed “ Strawberry ” .
This was interior document refresh by the issue show that project strawberry is about enable ai to do “ mysterious inquiry ” autonomously on the www .
When ask about the projection , OpenAI ’s representative read , “ these system will meliorate in logical thinking over meter .
” Now , what precisely is Project Strawberry and what precisely is OpenAI cookery ?
Is it the approaching GPT-5 modelling , aka ChatGPT 5 ?
allow ’s talk about it .
OpenAI ’s Project Strawberry excuse
Project Strawberry is say to berelated to Q * — the next find in AI maturation consider by OpenAI staff member inside the society .
There is not much selective information about Q * in the public land .
This was however , source who catch the demonstration say that it ’s up to of puzzle out skill and maths - associate dubiousness that no other current model can exhibit .
It has rack up over 90 % on a MATH dataset , a stringent bench mark for examination technique in maths problem .
This was bloombergrecently report that openai carry an national demonstration , where it shew a enquiry task that showcased homo - corresponding abstract thought .
OpenAI also gift dissimilar floor of AI forward motion , which we have depict in ourAGI explainer , and are as stick to :
dive into Project Strawberry
Project Strawberry is say to berelated to Q * — the next find in AI growth believe by OpenAI staffer inside the caller .
There is not much data about Q * in the public world .
However , source who look at the demonstration say that it ’s equal to of work scientific discipline and maths - tie in question that no other current model can present .
It has score over 90 % on a MATH dataset , a stringent bench mark for examination technique in maths job .
Bloombergrecently report that OpenAI support an national demonstration , where it show a inquiry projection that showcased human being - similar abstract thought .
OpenAI also present unlike level of AI onward motion , which we have describe in ourAGI explainer , and are as comply :
Many consider that with Project Strawberry , OpenAIhas reach Level 2where AI organization can argue intelligently like homo .
Reuters ’ story suggest that Project Strawberry include a post - training method acting where the mannequin is educate in a specific way of life , interchangeable to very well - tuning .
This specific proficiency for post - education is very like to Self - Taught Reasoner or STaR ( arXiv ) , a string - of - mean method acting , project by Stanford research worker in 2022 .
It take to ameliorate the modeling operation “ on complex abstract thought project like math or commonsensical dubiousness - answering ” by “ generate whole tone - by - footprint “ concatenation - of - persuasion ” rationale .
”
This was in mere term , star fundamentally permit the mannikin toimprove itself by hear from its own give logical thinking .
It seems that OpenAI is potential using the STaR proficiency for Project Strawberry to mime human - storey logical thinking and figure out strong mathematics problem .
dive into Project Strawberry
This specific proficiency for post - breeding is very like to Self - Taught Reasoner or STaR ( arXiv ) , a concatenation - of - conceive method acting , declare oneself by Stanford investigator in 2022 .
It place to ameliorate the good example functioning “ on complex abstract thought job like math or commonsensible dubiousness - answering ” by “ sire footmark - by - dance step “ strand - of - persuasion ” rationale .
”
In uncomplicated term , STaR fundamentally permit the theoretical account toimprove itself by memorise from its own beget abstract thought .
It seems that OpenAI is potential using the STaR proficiency for Project Strawberry to mime human - spirit level logical thinking and work concentrated maths job .
Project Strawberry is not just limit to innovative logical thinking , but it can also do something anticipate prospicient - visible horizon Tasks ( LHT ) , allot to written document view by Reuters .
fundamentally , it can also pretend as an agentive role and pasture the entanglement autonomously , amount up with finding , architectural plan , and do a serial of natural process .
This was ## the curious case of project strawberry on disco biscuit
on august 7 , openai ceo sam altman share an persona of strawberry mark and it parent fervour and rarity among the residential district .
This was the next twenty-four hour period , an cristal accounting ( @iruletheworldmo ) , conceive to be an openai staff member , twirp “ welcome to pull down two .
This was how do you experience ?
did I make you find ?
” Altman remark , “ awful tbh .
“
This was is chatgpt 5 finally issue forth ?
calculate at the ontogenesis so far , it look OpenAI is prepare to put out the next frontier modeling middling before long .
This was power by project strawberry , the frontier framework is probable to lend forward-looking logical thinking potentiality and unlock agentic workflow .
Now whether it will be call GPT-5 or something else , we do n’t eff .
Sam Altman has tell before that the replacement to the GPT-4 mannikin might not be call GPT-5 .
