Topics
Latest
AI
Amazon
Image Credits:Bryce Durbin / TechCrunch
Apps
Biotech & Health
Climate
Cloud Computing
Commerce
Crypto
initiative
EVs
Fintech
fundraise
Gadgets
Gaming
Government & Policy
ironware
layoff
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
Security
Social
Space
Startups
TikTok
Transportation
Venture
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
video
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
Has OpenAI invented an AI engineering science with the potential drop to “ threaten humanity ” ? From some of the late headline , you might be inclined to think so .
ReutersandThe Informationfirst describe last hebdomad that several OpenAI stave member had , in a alphabetic character to the AI startup ’s board of directors , flagged the “ artistry ” and “ likely peril ” of an internal research undertaking known as “ Q * . ” This AI project , according to the coverage , could solve certain mathematics problems — albeit only at form - school level — but had in the researchers ’ opinion a chance of build toward an elusive technological breakthrough .
There ’s now debate as to whether OpenAI ’s board ever meet such a varsity letter — The Verge cite asourcesuggesting that it did n’t . But the framing of Q * aside , Q * in actuality might not be as monumental — or threaten — as it sounds . It might not even be new .
AI researcher on X ( formerly Twitter ) , including Meta ’s chief AI scientist Yann LeCun , were directly skeptical that Q * was anything more than an extension of existing work at OpenAI — and other AI enquiry labs besides . In a Emily Price Post on X , Rick Lamers , who writes the Substack newssheet encipher with Intelligence , pointedto an MIT Edgar Guest lecture OpenAI co - founder John Schulman gave seven year ago during which he key a numerical mapping call “ Q * . ”
Several researchers conceive the “ Q ” in the name “ Q * ” refers to “ Q - erudition , ” an AI proficiency that serve a simulation learn and ameliorate at a particular task by taking — and being honour for — specific “ right ” actions . researcher say the star , meanwhile , could be a reference to A * , an algorithm for checking the nodes that make up a graphical record and explore the routes between these node .
Please ignore the deluge of complete meaninglessness about Q*.One of the principal challenges to meliorate LLM reliability is to exchange Auto - Regressive token prediction with provision . Pretty much every top research lab ( FAIR , DeepMind , OpenAI etc ) is working on that and some have already published …
Both have been around a while .
Google DeepMind applied Q - encyclopaedism to build an AI algorithm that could play Atari 2600 games at human point … in 2014 . A * has its origins in an pedantic newspaper published in 1968 . And research worker at UC Irvine several class agoexploredimproving A * with Q - eruditeness — which might be exactly what OpenAI ’s now pursuing .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
Nathan Lambert , a research scientist at the Allen Institute for AI , severalize TechCrunch he believes that Q * is associate to approaches in AI “ mostly [ for ] studying high school maths job ” — not destroying humankind .
“ OpenAI even shared work earlier this year meliorate the mathematical reasoning of language models with a technique predict process reward models , ” Lambert say , “ but what remains to be seen is how better mathematics abilities do anything other than make [ OpenAI ’s AI - powered chatbot]ChatGPTa salutary code assistant . ”
Mark Riedl , a computer skill prof at Georgia Tech , was likewise vital of Reuters ’ and The Information ’s coverage on Q * — and the broader media narrative around OpenAI and its pursuit toward contrived general intelligence ( i.e. AI that can do any undertaking as well as a human can ) . Reuters , citing a origin , implied that Q * could be a step toward artificial general news ( AGI ) . But researchers — admit Riedl — quarrel this .
citizenry are order this is Q*https://t.co / rI5dz0Maf7Seems plausible , though it ’s from May 2023 and no one lost their minds over this , nor should they .
“ There ’s no grounds that suggests that large language model [ like ChatGPT ] or any other engineering science under ontogeny at OpenAI are on a route to AGI or any of the day of reckoning scenario , ” Riedl told TechCrunch . “ OpenAI itself has at best been a ‘ fast follower , ’ having submit existing ideas … and discover ways to scale them up . While OpenAI hires top - charge per unit researchers , much of what they ’ve done can be done by research worker at other organizations . It could also be done if OpenAI investigator were at a dissimilar establishment . ”
Riedl , like Lambert , did n’t estimate at whether Q * might imply Q - encyclopedism or A * . But if it involved either — or a compounding of the two — it ’d be consistent with the current trends in AI research , he say .
“ These are all idea being actively act on by other research worker across academia and industry , with dozens of papers on these topics in the last six month or more , ” Riedl added . “ It ’s unlikely that researchers at OpenAI have had ideas that have not also been had by the substantive number of researchers also pursuing overture in AI . ”
That ’s not to suggest that Q * — which reportedly had the involvement of Ilya Sutskever , OpenAI ’s main scientist — might not move the needle forward .
Lamers verify that , if Q * uses some of the technique described in apaperpublished by OpenAI researchers in May , it could “ significantly ” increase the capabilities of language models . base on the report , OpenAI might ’ve happen upon a way to control the “ logical thinking chains ” of language models , Lamers says — enable them to guide models to accompany more desirable and logically sound “ paths ” to reach outcomes .
“ This would make it less likely that models follow ‘ foreign to human thinking ’ and misbegotten - patterns to reach malicious or wrong conclusions , ” Lamers say . “ I think this is really a win for OpenAI in condition of alignment … Most AI researchers agree we call for better ways to train these gravid model , such that they can more efficiently take info . ”
But whatever emerge of Q * , it — and the comparatively simple math equations it solves — wo n’t spell doom for humanity .
