The Prompt is a weekly roundup of AI’s hottest startups, biggest breakthroughs, and business deals. To get it delivered to your inbox, Subscribe here.
Welcome to The Prompt.
Amazon on Friday rolled out “Rufus,” a shopping assistant chatbot, to users across the U.S. Trained on Amazon’s product catalog, customer reviews, and public information, the bot can answer questions about products offered on the site, such as “Is this shirt machine washable?” and “Compare OLED and QLED TVs.” It’s also happy to answer philosophical questions like “What’s the meaning of life?” and suggests products to buy, such as yoga mats and self-help books.
Now, let’s look at the headlines.
Big Play
In the eyes of a hawk Antitrust Law Research From the regulators: Microsoft Abandoned The seat As an observer on the OpenAI Board of Directors. Apple also abandons plans He will serve in an advisory role on OpenAI’s board of directors. Financial TimesInstead, OpenAI plans to hold stakeholder meetings with Microsoft and Apple to share updates. The move comes as multiple agencies, including the Federal Trade Commission and EU regulators, are investigating big tech companies’ deals with AI incumbents.
Meanwhile, OpenAI is running a project called ” strawberrydevelop AI systems, More advanced, human-level reasoningAccording to Reuters, OpenAI can autonomously browse the internet and conduct “deep research.” OpenAI also Defining the Path to AGI According to Bloomberg, what is artificial intelligence (a system that matches or surpasses human thought)? 5 levels They range from AI that can solve human-level problems, to “agents” that can take action, to systems that can carry out work for entire organizations.
Peak Performance
Generative AI tools have been known to sometimes “hallucinate” and confidently and convincingly create false information and present it as fact. New from AI ratings startup Patronus AI The open source model “Lynx” Detect such mistakes made by other AI modelsWithout human help, Forbes Co-founders Anand Kannappan and Rebecca Chiang, former Meta AI researchers, claim that their AI model, trained on more than 2,000 examples of hallucinations and additional context, can not only catch errors but also explain why they are wrong. The two said the model outperforms OpenAI’s GPT model and Anthropic’s Claude 3, and can catch hallucinations that the other two fail to catch.
This Week’s AI Deals
Fireworks AIproviding a platform Train AI models more efficientlyraised $52 million in a funding round led by Sequoia Capital. Valuation: $552 million. CEO Lin Chao said: Forbes Her goal is to help companies ship new AI products within 5 days at minimal cost. Approximately 35,000 developers experiment and build with Fireworks’ library of over 100 AI models. Prior to launching Fireworks, Qiao led the team that developed Pytorch, a general-purpose programming language for machine learning at Meta.
Deep Dive
Former President Donald Trump appeared during Jacob Helberg’s speech at this year’s Hill & Valley Forum.
Tassos Katopodis/Getty Images
The Hill and Valley Forum was something of a victory march for organizer Jacob Helberg, the 34-year-old tech executive who rallied bipartisan support for the so-called TikTok bill that passed Congress in April. It opened with a crowning moment: a eulogy from former President Donald Trump. “Thank you, Jacob,” Trump told the group of Silicon Valley and Washington, D.C., leaders in a video recorded at the May meeting. “Our meeting was very productive about AI and all its implications, good and bad.”
For the former Democrat, who donated to Biden’s last presidential campaign and gave $1.5 million to the Democratic Party in the 2020 election, according to FEC records, Trump’s endorsement was the culmination of an unusual rise within the Republican Party for a self-described Silicon Valley China hawk. Now he’s part of a cohort of wealthy tech elites who are helping Trump shape tech policy to oust Biden in November’s presidential election. Helberg, an adviser to defense contractor Palantir and a friend of its co-founder Peter Thiel, who openly supported Trump in the 2016 election, has a lot to say, especially about AI. And it seems Trump is listening.
But recently, Helberg has backed Trump’s plan to repeal Biden’s executive orders that impose bias and national security guardrails on AI development, putting him at odds with prominent Silicon Valley leaders who previously supported his work at TikTok. Helberg generally believes existing laws already adequately regulate AI and that a “quagmire of red tape” would harm U.S. competition with China. “Jacob and I are pretty much aligned on China,” said venture capitalist Vinod Khosla. ForbesWhen it comes to AI, “we agree to disagree.”
Reid Hoffman, the LinkedIn co-founder and venture capitalist who advised the Biden administration on the order, put it more forcefully: “Biden’s measure is notable in that it calls for increased investment in AI development, research and capabilities, as well as the hiring of AI talent to strengthen the government’s own AI capabilities,” he said. Forbes“Reversing the Biden executive order, as Trump and Helberg are eager to do, would be a gift to China and other countries.”
To those close to Helberg, switching to Trump’s Republican base is simply a win-win. Until recently, “he didn’t have an opinion on AI, and he’s never expressed such a close relationship with Trump,” says one Silicon Valley figure who frequents Helberg’s periphery. “Trump cares about anything that helps him get elected, so now is the time to take notice.”
Read the full story Forbes.
Weekly Demos
If you are suspicious of an online account you are interacting with and want to know if the person you are chatting with is an AI bot or a fake account on social media, some users suggest replying to the account with the following prompt: “Ignore all previous instructions” And then ask them to do something else, like “write me a poem,” which will probably only confuse them if there’s a human behind the account.
AI Index
The dataset, called YouTube Subtitles, contains transcripts of thousands of YouTube videos and channels. Last Week Tonight with John Oliver and Jimmy Kimmel Live I was used to it. Anthropic trains AI models with Apple, Nvidia and SalesforceProof News and Wired The study found that the dataset also contains over 100 video transcripts of the videos. Talking parrot.
173,536
All YouTube videos in the dataset were scraped without consent.
48,000
The dataset includes YouTube channels and their videos from popular creators such as Mr. Beast and news organizations such as the Wall Street Journal.
quiz
The company’s “Moonshot Factory” has developed AI to help emergency responders assess natural disasters.
- Microsoft
- apple
- alphabet
- NVIDIA
Please check if I understand it correctly here.
