AI's first language is English — a bias that researchers are racing to counter before it gets permanently baked into the new technology, Axios' Alison Snyder reports.

Why it matters: Most of today's generative AI tools are built on large language models (LLMs) trained on texts and data in English and Chinese, leaving the 6 billion native speakers of the world's more than 7,000 other languages at risk of being left out as the technology reframes work, business, education, art and more.

• "The languages that these models serve are going to be defaulted to as the easiest way to generate new information," says Sara Hooker, who leads Cohere for AI, Cohere's nonprofit AI research lab.

How it works: Most of the data used to train the foundational models fueling the current wave of AI — for example, OpenAI's GPT models and Meta's Llama versions — is in English, and the AI tools they support perform best when asked questions in it.

• GPT-4, the latest LLM from OpenAI, excels at English, Spanish, Italian, Indonesian and other Latin alphabet-based languages, but it struggles with Thai, Punjabi and other languages based on different alphabets. Baidu's Ernie Bot is best with Chinese, which it was trained on.
• When it released its updated LLM model in July, Meta cautioned that because most of the training data for the model is in English, it "may not be suitable for use in other languages."
• ChatGPT can translate prompts and responses into English well, but it often fumbles translating English into other languages.
• ChatGPT can also make up words, struggle with syntax and generate gibberish in many underrepresented languages, Andrew Deck writes for Rest of World, which tested the abilities of the free version of the chatbot released late last year.

What's happening: Some developers are trying to overcome these linguistic shortcomings by focusing on building multilingual large language models, while others are putting their efforts into tuning models to a particular language.

• The Aya Project at Cohere is an open science project to build an AI model tuned with instructions in 100 languages (rather than focusing on a foundational model trained on unstructured text). Aya, which plans to release its model early next year, follows other open-source models, including BLOOM, which can generate text in 46 languages.

Clibrain — a startup based in Madrid — in July released LINCE Zero, an LLM tuned to Spanish.

• Clibrain is focused on capturing the nuances of the language, which has numerous dialects and variations spoken in 20 countries around the world.
• "Spanish is not one Spanish," the company's CEO, Elena González-Blanco, tells Axios.

Yes, but: Building a model for every language isn't realistic, says Mona Diab, director of the Language Technologies Institute at Carnegie Mellon University. She's a proponent of multilingual models and sees promise in using them to capture families of languages.

• For example, a model trained on Arabic from Tunisia, Egypt and Saudi Arabia, but not from Qatar, may still be able to respond to a prompt in Qatari Arabic dialect.

What to watch: Researchers working on the Aya Project are now red-teaming eight of the languages in the model for safety, biases and other risks.

• Native speakers are annotating the responses for toxicity, unsafe use, the prompting of bad financial advice, and other issues.
• "All the red-teaming to date for these major model launches has been done primarily in English," Hooker says, adding that how to assess safety is an open research question.
• "We report risk in just one language, but we're deploying technology all over the world."

A prominent civil rights group is launching a center to investigate how artificial intelligence affects civil rights, fosters racism and spreads bigotry, Axios Latino's Russell Contreras reports.

Why it matters: The center arrives amid rising concern that AI may fuel racism and more antisemitism in the U.S. by amplifying bias from human-generated content on the internet.

Details: The Leadership Conference on Civil & Human Rights (LCCHR), one of the nation's largest and oldest civil rights coalitions, announced Thursday the creation of the Center for Civil Rights and Technology.

• The center will monitor legislation and regulations on AI and other emerging technologies and assess how those will impact civil and human rights.

State of play: Today's leaders in AI, companies like OpenAI, Google and Microsoft, have aimed to filter out the most egregious racism, sexism and homophobia.

Yes, but: Others, waving free-speech or "anti-woke" banners, are already gearing up to create similarly robust systems without such limits.

What they're saying: "Hate has had a very comfortable home on our social media platforms. And it has been given cover by, sadly, sometimes, by leaders that have every ability to disrupt that," LCCHR CEO Maya Wiley told Axios.

But, but, but: Wiley said the center will also highlight examples of how AI can be used to understand civil rights better.

• Museums that focus on racial violence and antisemitism have begun using holograms, artificial intelligence and virtual reality to confront difficult episodes in history.

C3.ai is promising to deliver fine-tuned generative AI to any business in 12 weeks — addressing a problem faced by businesses which have struggled to organize their data and gotten stuck in pilot stages with their generative AI plans, Axios' Ryan Heath reports.

Driving the news: C3.ai launched 28 domain-specific generative AI products Wednesday, offering smaller AI models than those used by the most popular AI chatbots.

Why it matters: Generative AI tools need to meet specific needs to be game-changers for business customers — including achieving compliance with regulations and internal standards, and delivering answers with certainty.

What's happening: C3.ai says its products provide the same answer every time if given the same input — eliminating the hallucinations found in large language models — and that all answers from its products "are immediately traceable with one click."

• The company said it has already built out generative AI products for customers ranging from the Missile Defense Agency to Pantaleon, a Guatemala-based sugar company.

The details: The price tag to get a C3.ai model up and running is $250,000 annually, plus a fee per CPU hour.

What they're saying: "We spent 15 years and a couple of billion dollars for software engineering getting ready for this," C3 CEO Tom Siebel said of his pitch for organizations to embrace "small language models."

• "We don't expect this debate to be settled until experiments and in-the-field usage shows where each type of model excels or fails," Fred Havemeyer, Macquarie senior enterprise software analyst, told Axios, adding that hybrid models may win in the end.

Yes, but: C3.ai's share price dropped around 12% Thursday, after the firm announced earnings in line with analyst expectations.

• Investors signaled they cared more that the company would not reach profitability by the end of fiscal year 2024, as it previously forecast, than about its new products.

• Anthropic launched a paid version of its Claude chatbot. (The Verge)
• EBay is using AI to allow new sellers to create listings from just a photo. (Techcrunch)
• Your car could be delivering even more of your personal data to corporations than your clicks online, per a new report. (Axios)
• The Senate confirmed Anna Gomez as FCC commissioner, delivering a Democratic majority to the agency. (Axios Pro)
• Trading places: KPMG announced a new AI and digital innovation group, led by Steve Chase, who will join CEO Paul Knopp's leadership team.

Here's to the next generation of journalists.