Topics
a la mode
AI
Amazon
Image Credits:William Tunstall-Pedoe, founder, Unlikely AI (© 2017 Yolande De Vries)
Apps
Biotech & Health
Climate
Image Credits:William Tunstall-Pedoe, founder, Unlikely AI (© 2017 Yolande De Vries)
Cloud Computing
Commerce
Crypto
Enterprise
EVs
Fintech
Fundraising
gizmo
Gaming
Government & Policy
Hardware
layoff
Media & Entertainment
Meta
Microsoft
privateness
Robotics
Security
Social
place
Startups
TikTok
Transportation
Venture
More from TechCrunch
event
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
After announcing a whopping $ 20 million seed last class , Unlikely AIfounder William Tunstall - Pedoe has hold the bud U.K. foundation role model manufacturer ’s approach under lock and key . Until now : TechCrunch can exclusively reveal Unlikely is taking a “ neuro - symbolical ” approach to its AI . In an additional development , it ’s annunciate two senior hires — including the former CTO of Stability AI , Tom Mason .
Neuro - symbolic AIis a type of stilted word that , as the name suggests , integrates both the modern neural meshing approaches to AI — as used by large language modelling ( LLMs ) , like OpenAI ’s GPT — and earlier Symbolic AI architectures to plow the weaknesses of each .
Tunstall - Pedoe gain public profile in the U.K. technical school scene back in 2012 whenAmazon acquiredhis voice helper startup , Evi . Two years later Amazon launched the Echo and Alexa , incorporate much of Evi ’s engineering science . With Unlikely AI , Tunstall - Pedoe is direct to put himself back in the spotlight as he takes the wraps off the technology he and his team have been exercise on since 2019 , when the inauguration was founded .
At Stability AI , meanwhile , Mason manage the development of major foundational models across various battleground and helped the AI company put forward more than $ 170 million . Now he ’s CTO of Unlikely AI , where he will oversee its “ symbolic / algorithmic ” approach .
In addition , Fred Becker is joining as chief administrative officer . He previously concord senior part at company let in Skype and Symphony . At Unlikely , his role will be to shepherd its now 60 full - prison term stave — who are based for the most part between Cambridge ( U.K. ) and London .
The AI startup claims its attack to foundational AI models will endeavor to avoid the risks we ’ve rapidly become all - too - familiar with — namely predetermine , “ hallucination ” ( aka fabrication ) , accuracy and trust . It also claims its approach will use less energy in a bid to reduce the environmental impact of Big AI .
“ We ’ve been working privately for a number of years and we ’re very excited about our two fresh senior hire , ” Tunstall - Pedoe secern TechCrunch over a call .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
Fleshing out the squad ’s approach , he went on : “ We ’re building a ‘ trustworthy ’ AI platform that ’s designed to plow middling much all of the key issue with AI at the moment , as it pertains to … hallucination and accuracy . We ’re combining the capabilities of generative AI , statistical AI , with emblematic algorithmic methods , [ and ] conventional software method to get expandability and dependableness . ”
He described the program as “ horizontal ” in that it would “ compound many dissimilar eccentric of practical app . ”
Of the exact software , he was more coy — but continue to stress the phrase “ trustworthy AI . ”
For his part , Mason said his sentence at Stability AI saw the company work up “ some amazing models ” and “ an incredible ecosystem around the models and the technology , ” as he put it . It also have theabrupt exitof father Emad Mostaque , followed by a routine of other gamey - visibility team departures . While Mason care his former colleague “ all the best , ” he said said he ’s “ super aroused ” to join Unlikely AI .
Digging into the inauguration ’s technology , Tunstall - Pedoe said the platform is write of two things : “ The word ‘ neuro ’ and the word ‘ symbolic . ’ ‘ Neuro ’ implies deep learning , so solving problems that machine have not been able to solve for decennary … ‘ Symbolic ’ refers to the kind of software that powers your spreadsheet or other applications .
“ One of the weaknesses of ‘ neuro ’ is that it ’s sometimes awry . When you cultivate a example , you give it datum , it gets good and better . But it never gets to 100 % . It ’s right , for case , 80 % of the time , which means it ’s wrong 20 % of the time . ”
He sound out this is “ fabulously damaging to trust ” because “ the neuro calculation is opaque . ” Indeed , there ’s an entire field of enquiry render to sympathise what happen inside these huge LLMs .
Instead , he pronounce unbelievable plans to flux the certainties of traditional software , such as spreadsheet , where the calculations are 100 % precise , with the “ neuro ” approach in generative AI .
“ What we ’re doing is combine the best of both worlds , ” suggested Tunstall - Pedoe . “ We ’re taking the capabilities of LLMs , of all the progression in deep learning , and we ’re combine it with the trustiness and expandability and other advantages — admit things like price and environmental encroachment — of non - statistical machine learning … The sight we have of AI is all of those capabilities , but in a fashion that ’s entirely trustworthy . ”
He argue a combined approach shot will bring cost and environmental benefit , too , compare to today ’s LLMs : “ These models are incredibly expensive [ to go ] and environmentally unfriendly , but they are also pricey in term of faith by make answers that are haywire . ”
Why have n’t other foundational models film a similar route ?
“ I consider that that ’s happening , ” Mason react . “ Sometimes we talk about it as ‘ compound architecture . ’ We ’ve seen the rise of thing likeRAG . That ’s a kind of compound architecture . This is very much in the same vein , but it ’s building on all of that with the vantage symbolic logical thinking , making it possible to have wholly precise abstract thought . ”
In this respect , he said Mason believes Unlikely AI is “ ahead of the undulation . ”
Another dubiousness is whether Unlikely AI will grow a fuller foundational model , such as OpenAI — or take a interracial approaching , akin to Mistral ’s , offering both foundational and open source model ?
Tunstall - Pedoe order the company is still yet to make up one’s mind the direction of travel : “ We have n’t made any conclusion like that yet . That ’s part of internal discussions . But we ’re work up a platform and the relief is TBD … It ’s a decision that we ’re going to make in the skinny future . ”
One thing is sustain , though : It ’s going to be ramp up out of London and Cambridge : “ Obviously we ’ve make a much pocket-sized population than in the U.S. and China . But London is a wonderful property to be build an innovative AI startup . There ’s lots of talent here . caboodle of innovation . ”
While the fashion model discharge timeline is n’t clear , Unlikely AI is sure about the strength of its ambition . Given AI is the number one strategic priority of every trillion - dollar mark market crown company out there , Tunstall - Pedoe enounce he ’s tear for major adoption . “ We want to be massively successful , we require to have a vast impact . We ’re certainly subject to unlike path of achieving that , ” he add .