Google Foto's AI kan gorilla's nog steeds niet labelen

Google Foto's AI kan gorilla's nog steeds niet labelen

Bronknooppunt: 2685385

AI in het kort AI may be progressing rapidly, but it appears Google still hasn't quite fixed an eight-year problem with its image recognition system: identifying pictures of gorillas accurately.

The company was criticized when a software developer, Jacky Alciné, found the image recognition system deployed in its Photos app in 2015 had mistakenly labelled a photo of him and his friend as gorillas. Alciné and his friend are both Black. Horrified, Google swiftly covered up the racist error by blocking the software's ability to label images of gorillas.

Eight years later, the flaw still hasn't been fixed properly. An experiment uitgevoerd door de New York Times ontdekte dat gebruikers hun foto's konden doorzoeken op verschillende soorten dieren, zoals katten en kangoeroes, maar niet op gorilla's, bavianen, chimpansees, orang-oetans en apen. 

The Photos app still steers clear of labelling images of primates. It's not clear whether the company is unable to fix the issue, or if the feature is too controversial or risky to deploy. Interestingly, Apple's Photo app and Microsoft's OneDrive storage app had the same glitch too, while Amazon Photos tended to mistakenly label other types of animals as gorillas.

Alciné said he was disappointed Google has not fixed the problem. "I'm going to forever have no faith in this AI," he said.

OpenAI lanceert een nieuw programma om ideeën te financieren over hoe AGI democratisch te controleren

De non-profittak van OpenAI zegde in totaal een miljoen dollar toe aan mensen met goede ideeën over hoe een democratisch proces kan worden opgezet waarbij burgers de ontwikkeling van AI kunnen vormgeven.

Iedereen kan zijn voorstellen indienen tot 2 juni. De AI-startup selecteert de tien beste aanvragen en kent elk $ 100,000 toe, zodat de ontvangers kunnen beginnen met het bouwen van prototypen van hun ideeën. 

"AGI should benefit all of humanity and be shaped to be as inclusive as possible," OpenAI aangekondigd in a blog post this week. "We are launching this grant program to take a first step in this direction. We are seeking teams from across the world to develop proof-of-concepts for a democratic process that could answer questions about what rules AI systems should follow."

OpenAI said that the grants are experiments, and it may or may not apply any of the ideas into practice. The startup is interested in methods that support input from a diverse community on problems that can't necessarily be solved by following laws.

"For example: under what conditions should AI systems condemn or criticize public figures, given different opinions across groups regarding those figures? How should disputed views be represented in AI outputs? Should AI by default reflect the persona of a median individual in the world, the user's country, the user's demographic, or something entirely different? No single individual, company, or even country should dictate these decisions," it asked.

Anthropic haalt 450 miljoen dollar op in Series C-ronde

Anthropic, een AI-startup die mede is opgericht door ex-OpenAI-medewerkers, heeft $ 450 miljoen opgehaald in zijn laatste investeringsronde onder leiding van Spark Capital om meer commerciële producten te bouwen.

De Series C-ronde wordt ondersteund door Google, Salesforce Ventures, Sound Ventures, Zoom Ventures en anderen. Anthropic werkt momenteel aan het werven van meer klanten met behulp van zijn grote taalmodel, Claude. CEO Dario Amodei zei dat het geld zal gaan naar het bouwen van meer producten.

"We are thrilled that these leading investors and technology companies are supporting Anthropic's mission: AI research and products that put safety at the frontier. The systems we are building are being designed to provide reliable AI services that can positively impact businesses and consumers now and in the future."

Anthropic said it was focused on building safer products using AI alignment techniques to guide its chatbot into following instructions and generating appropriate responses. The company recently expanded Claude's contextvenster om 100,000 tokens te ondersteunen, waardoor het honderden pagina's aan documenten in één keer kan verwerken. 

Waymo werkt samen met Uber om zijn robotaxi-service uit te breiden

Zelfrijdende auto biz Waymo werkt samen met de beste taxi-app Uber om meer passagiers aan te trekken in Phoenix, Arizona. De twee streden op een gegeven moment fel tegen elkaar, maar lanceren nu op hun eigen manier autonome autotechnologie en zien de voordelen van samenwerking.

Waymo zal zijn zelfrijdende voertuigen leveren, terwijl Uber zijn netwerk van rijders zal leveren. Uber-gebruikers kunnen Waymo Driver-auto's aanhouden om eten te bezorgen dat is besteld via de Uber Eats-app, of om via de Uber-app naar hun doelbestemmingen te worden gereden.

"We're excited to offer another way for people to experience the enjoyable and life-saving benefits of full autonomy," Tekedra Mawakana, co-CEO of Waymo, uitgelegd in a statement this week. "Uber has long been a leader in human-operated ridesharing, and the pairing of our pioneering technology and all-electric fleet with their customer network provides Waymo with an opportunity to reach even more people."

De robotaxi-service wordt naar verwachting later dit jaar gelanceerd in Phoenix, Arizona. ®

Tijdstempel:

Meer van Het register