- AI Odyssey
- Posts
- AI language problem 👀
Brought to you by
Hello there! 👋 Welcome back to AI Creator. Get ready for a dose of exciting AI updates and intriguing reads. Just scroll down to dive in!
⏰ Top 5 Headlines
🔗Link: AI Grant second batch announced. The increasingly popular and successful AI grant has announced its second cohort of start-ups.
🔗Link: IRS will use AI to crack down on wealthy potential tax violators.
🔗Link: Morgan Stanley to launch AI chatbot to woo wealthy.
🔗Link: AI Drake song ‘not eligible’ for a Grammy, Academy CEO says.
🔗Link: Google will require political ads that use A.I. to alter imagery or sounds to be prominently marked.
Today's AI Creator issue is brought to you by Pangea
Pangea AI was built by a team of YC founders with years of experience helping startups find the right talent.
Just write a job title and it generates a custom and well-scoped JD to help you attract the right talent. Our AI will help you identify pre-vetted, highly relevant candidates from our private network.
🚀 Company Spotlight
💬 Tool of the Week
🔗Snipo: A side-by-side AI notetaker for your videos.
💬 Longer reads
Major AI product updates were announced by HubSpot, Slack, and Intuit:
HubSpot introduces new AI tools for marketing and sales, including AI content assistants, automated customer service AI agents, AI insights for customer behavior prediction, and ChatSpot for content inquiries.
Slack adds AI features such as thread and discussion summarization, intelligent chat and file search, and project tracking lists.
Intuit unveils "Intuit Assist," an AI-powered financial assistant, serving TurboTax, Credit Karma, QuickBooks, and MailChimp users, offering support with tax-related queries, financial guidance, expense monitoring, invoicing, email campaigns, and more. Aimed at assisting small businesses and individuals.
AI's language gap
Illustration: Natalie Peeples/Axios
The bias in AI language models, which mainly focus on English and Chinese, excludes billions of speakers of 7,000+ other languages, affecting work, education, and more. Researchers are countering this bias with multilingual and language-specific models, like Cohere's Aya Project, Inception's Arabic-English model, and Clibrain's LINCE Zero for Spanish. Advocates suggest grouping languages into families to bridge gaps. This bias also underscores the need for culturally aware training data, considering the histories and politics of various cultures. Researchers are evaluating language model responses, addressing safety concerns, and biases. Overall, rectifying the English-centric bias is crucial for an inclusive and culturally sensitive AI. 🔗Read more
Apple is reportedly spending “millions of dollars a day” training AI
Apple is dedicating millions of dollars daily to advance its artificial intelligence efforts. The company is actively working on various AI models across different teams, with a particular focus on conversational AI through its "Foundational Models" unit, led by John Giannandrea, Apple's head of AI.
Diverse AI initiatives: Apple is not limited to just one AI project. Besides conversational AI, it has teams dedicated to visual intelligence and multimodal AI, which can understand and generate images, videos, and text. These AI models are aimed at enhancing user interactions and automating tasks, including customer support and Siri functions.
Powerful AI language model Ajax GPT: Apple's most advanced language model, internally referred to as Ajax GPT, has been trained on over 200 billion parameters. It is reportedly more powerful than OpenAI's GPT-3.5, which underpins ChatGPT. This model, initially developed for internal use, is highly confidential within the company. These developments highlight Apple's significant commitment to artificial intelligence and its aspirations to be at the forefront of AI technology. 🔗Read more
Generative AI could revolutionize email—for hackers
Generative AI, including OpenAI's ChatGPT, is transforming the landscape of email security, posing a significant challenge to phishing defenses. Traditionally, phishing emails often displayed red flags such as poor grammar and spelling errors, but AI-powered tools can now effortlessly eliminate these markers. This technology allows hackers, even those with limited expertise, to craft highly personalized and convincing phishing emails within seconds, making it difficult for recipients to distinguish them from legitimate messages. Email security experts have observed instances of generative AI-generated phishing emails that are so well-crafted that they escape initial detection. As the capabilities of AI continue to evolve, the future of email security hinges on the development of defensive AI systems and enhanced employee training to counter these emerging threats. 🔗Read more
🤳 Extra reads
🔗Link: ChatGPT doesn’t want to write your Stucky fanfic.
🔗Link: Arm sees sales growth of more than 20% next year boosted by AI.
🔗Link: An ad consultant used AI to create a fashion magazine but found it nearly impossible to avoid 'too perfect' images.
🔗Link: 5 best books to read to get smart about AI.