Topics
tardy
AI
Amazon
Image Credits:Kimberly White/Getty Images for TechCrunch
Apps
Biotech & Health
Climate
Image Credits:Kimberly White/Getty Images for TechCrunch
Cloud Computing
Commerce Department
Crypto
Output from the Lucid v1 model.Image Credits:Rami Seid
endeavour
EVs
Fintech
Image Credits:Google DeepMind
fund-raise
Gadgets
back
Government & Policy
ironware
Layoffs
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
Security
Social
Space
Startups
TikTok
Transportation
Venture
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
On Monday , Anthropic CEO Dario Amodei pose in for a five - hour podcastinterviewwith AI influencer Lex Fridman . The two covered a image of topics , from timeline for superintelligence to progress on Anthropic ’s next flagship tech .
To spare you the download , we ’ve pulled out the salient point .
Despiteevidenceto the contrary , Amodei believe that “ surmount up ” models is still a workable path toward more capable AI . By scaling up , Amodei clarified that he mean increasing not only the amount of compute used to train simulation , but also models ’ size — and the size of it of model ’ training solidifying .
“ Probably , the scaling is going to continue , and there ’s some magic to it that we have n’t really explained on a theoretic basis yet , ” Amodei said .
Amodei also does n’t think a shortage of data point will present a challenge to AI growing , unlike someexperts . Either by generating synthetic data point or extrapolating out from exist data , AI developers will “ get around ” data limitations , he says . ( It remains to be seen whether theissueswith synthetic data are resolvable , I ’ll note here . )
Amodei doesacknowledge that AI compute is probable to become more costly in the near full term , partly as a result of scaling . He expects companies will spend one thousand million of dollars on clusters to train models next year , and that by 2027 , they ’ll be spending C of one million million . ( Indeed , OpenAI isrumoredto be planning a $ 100 billion data centre . )
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
And Amodei was plainspoken about how even the best models are unpredictable in nature .
“ It ’s just very hard to control the behaviour of a model — to steer the demeanour of a model in all circumstances at once , ” he said . “ There ’s this ‘ whack - a - mole ’ aspect , where you push on one thing and these other things start to move as well , that you may not even acknowledge or measuring rod . ”
Still , Amodei anticipate that Anthropic — or a rival — will make a “ superintelligent ” AI by 2026 or 2027 — one exceeding “ human - level ” performance on a issue of tasks . And he worries about the implications of this .
“ We are rapidly running out of really convincing blockers , in truth compelling understanding why this will not happen in the next few years , ” he say . “ I worry about economics and the concentration of power . That ’s in reality what I worry about more — the insult of power . ”
Good matter , then , that he ’s in a position to do something about it .
News
An AI newsworthiness app : AI newsreader Particle , launch by former Twitter applied scientist , aims to help readers better understand the tidings with the help of AI applied science .
author raises : Writer has raise $ 200 million at a $ 1.9 billion valuation to expand its enterprise - focused generative AI platform .
Build on Trainium : Amazon Web Services ( AWS ) has set up Build on Trainium , a new program that ’ll present $ 110 million to institutions , scientists , and students explore AI using AWS infrastructure .
Red Hat buys a inauguration : IBM ’s Red Hat is acquiring Neural Magic , a startup that optimize AI manikin to operate faster on trade good processor and GPUs .
Free Grok :X , formerly Twitter , is testing a loose version of its AI chatbot , Grok .
AI for the Grammy : The Beatles ’ path “ Now and Then , ” which was refined with the use of AI and released last year , has been nominated for two Grammy awards .
Anthropic for demurrer : Anthropic is team up with data analytics firm Palantir and AWS to supply U.S. word and defense federal agency access to Anthropic ’s Claude family of AI models .
A new domain : OpenAI bought Chat.com , adding to its assemblage of high - profile knowledge base gens .
Research paper of the week
Google claim to have develop an improved AI model for flood forecasting .
The model , which build on the company ’s previous work in this country , can prognosticate flooding shape accurately up to seven days in advance in lots of countries . In theory , the model can give a flood prognosis for anywhere on Earth , but Google note that many region lack diachronic data to validate against .
Google ’s offer a waitlist for API access to the manikin to disaster direction and hydrology experts . It ’s also making prognosis from the model useable through itsFlood Hubplatform .
“ By making our forecast usable globally on Flood Hub … we hope to contribute to the research residential district , ” the caller write in ablog post . “ These data can be used by expert users and research worker to inform more cogitation and analysis into how floods impact communities around the world . ”
Model of the week
Rami Seid , an AI developer , has resign a Minecraft - feign model that can play on a single Nvidia RTX 4090 .
like to AI startup Decart ’s of late released “ open - world”model , Seid ’s , call Lucid v1 , emulates Minecraft ’s plot mankind in real time ( or tight to it ) . Weighing in at 1 billion parameters , Lucid v1 get hold of in keyboard and computer mouse social movement and generates frames , sham all the physics and artwork .
Lucid v1 suffers from the same limitation as other game - simulate model . The solving is quite low , and it tends to quickly “ block ” the level layout — turn your graphic symbol around and you ’ll see a rearranged view .
But Seid and his partner , Ollin Boer Bohan , say they plan to continue develop the model , which is available fordownloadand magnate the online demohere .
Grab bag
DeepMind , Google ’s premier AI lab , has released the code forAlphaFold 3 , its AI - powered protein prediction model .
AlphaFold 3 was announced six months ago , but DeepMind polemically withheld the code . Instead , it provided access code via a World Wide Web server that restricted the number and types of predictions scientist could make .
Critics saw the move as an effort to protect DeepMind ’s commercial interests at the expense of reproducibility . DeepMind spin - off , Isomorphic Labs , is applying AlphaFold 3 , which can mould protein in concert with other mote , to drug discovery .
Now academician can use the framework to make any prevision they like — including how protein comport in the comportment of likely drugs . Scientists with an academic tie canrequest code access here .
