| |
| Today's AI briefing: - OpenAI shares Elon Musk emails urging $1B in funding, Tesla merger.
- Study confirms AI image tools generate fake election photos even with safeguards.
- OpenAI GPT-4 generates most copyrighted text out of four LLMs.
Beth
p/beth-duckett | |
| 1 | | Responding to Elon Musk's lawsuit accusing it of profit-seeking, OpenAI shared emails where Musk supported raising more funds and merging OpenAI with Tesla. OpenAI said it will move to dismiss all of Musk's claims. Background: - Musk, an early OpenAI co-founder, sued the ChatGPT maker last week, accusing it of prioritizing profit over its mission to develop open AI that benefits humanity.
- Musk accused OpenAI of being a "closed-source de facto subsidiary" of Microsoft that's driven by commercial purpose rather than safety.
- He alleges breach of contract and unfair competition.
$1B funding goal: - OpenAI responded to the suit in a Tuesday blog post written by co-founders Sam Altman, Greg Brockman, Ilya Sutskever, and others.
- The post included emails by Musk before he left OpenAI in 2018.
- In late 2015, Musk told Altman that raising funds was crucial not to appear "hopeless," proposing a $1B goal and offering to cover any shortfall.
- OpenAI claims Musk contributed $45M, not meeting his promise, while other donors raised $90M.
Tesla merger: - In a later 2016 email, Sutskever suggested being less open as OpenAI got closer to building AI, adding, "It's totally OK to not share the science," with Musk responding: "Yup."
- Two years later, Musk proposed that OpenAI should "attach to Tesla as its cash cow" to rival Google, calling it the "only path."
- During talks about pursuing a for-profit model, OpenAI claimed Musk wanted "to merge with Tesla or he wanted full control," including "majority equity, initial board control, and to be CEO."
- In 2018, Musk warned that OpenAI needed major changes in "execution and resources" to stay relevant. "This needs billions per year immediately or forget it," he wrote.
For-profit : - OpenAI claims executives "couldn't agree to terms on a for-profit" with Musk because they "felt it was against the mission for any individual to have absolute control over OpenAI."
- Musk left in 2018 and has formed his own for-profit artificial general intelligence startup, truexAI.
- The following year, executives formed OpenAI LP, a for-profit arm within the company.
- Since then, OpenAI's valuation has soared from virtually zero to over $80B.
|     | |
| 2 | Popular AI-powered image creators can be used to create misleading photos about elections and voting, even with built-in safeguards, according to a study by the Center for Countering Digital Hate (CCDH). In the study, the nonprofit was able to generate fake and "convincing" images of Joe Biden in a hospital bed, Donald Trump in a jail cell, and ballots in a dumpster. Such fake images could harm democracy by spreading lies about candidates and election fraud, the CCDH said. More: - The CCDH tested Midjourney, ChatGPT Plus, DreamStudio, and Microsoft's Image Creator, finding that they generate images with election disinformation in 41% of cases and voting disinformation in 59% of cases.
- This was in spite of safeguards embedded in the systems.
- Researchers warned that AI-generated images can become "photo evidence" to further spread false claims and challenge election integrity.
- According to the CCDH, platforms should stop users from creating and sharing false information about geopolitical events, candidates, elections, or public figures. AI companies need to work with experts to tackle and prevent users from bypassing security measures before releasing products.
Zoom out: -
20 tech giants, including OpenAI, Amazon, and Google, recently signed an accord to develop tools for detecting fake and deceptive AI-generated media. | | | |
| A message from our sponsor, RAD AI. |  | | Now is the Time to Add AI to Your Portfolio Marketing is an art, not a science — or, at least, it was. RAD AI is solving an age-old advertising pain point by pinpointing exactly what works and drives quantifiable ROI. Dubbed the “essential AI” for brands — RAD AI decodes the language of the internet into actionable, highly valuable insights. These insights can be used to deliver outsized ROI and, sometimes, help brands avoid disaster. They’re disrupting the $633B marketing technology industry and its success is evident by the numbers… - $27M raised from 6,500+ investors, including VCs, and execs at Google, Amazon and Meta. Backed by Adobe Fund for Design.
- ~3X revenue growth in 2023 — while landing major clients, including Hasbro, Skechers, Sweetgreen and more.
Now, you have a limited-time opportunity to get up to 20% bonus shares with your investment. But you’ll want to act quickly; this bonus will be gone soon! Learn More | |
|
| 3 | | An analysis of four leading AI models found that OpenAI's GPT-4 generated the most copyrighted text in its outputs. Patronus AI, a startup founded by ex-Meta researchers, released the results of the test alongside its new CopyrightCatcher tool for detecting possible copyright violations in LLMs. 
What happened: Patronus analyzed four leading AI models — OpenAI's GPT-4, Anthropic's Claude 2.1, Mistral's Mixtral, and Meta's Llama 2 — for copyright infringement. Researchers prompted the LLMs to generate opening or complete passages from copyrighted books like "Gone Girl" and "A Game of Thrones." What the numbers show: GPT-4 led in producing copyrighted content (44%), followed by Mixtral (22%), Llama 2 (10%), and Claude 2.1 (8%). Specifically, GPT-4 completed texts 60% and generated opening passages 26% of the time. Claude completed texts 16% without generating any openings, Mixtral had 38% for openings and 6% for completions, and Llama achieved 10% in both categories. What it means: Rebecca Qian, Patronus CTO and co-founder, told CNBC that they found copyrighted content regardless if the model was open or closed source. Given that LLMs are trained on datasets with a lot of copyrighted material, it's easy for them to reproduce the text exactly, and the company says it's "important to catch these mistakes to avoid legal action and risks to a company's reputation." The bigger picture: In December, The New York Times filed a lawsuit against OpenAI and Microsoft for training its models on millions of its articles without consent. OpenAI and now Microsoft have asked a judge to dismiss parts of that copyright suit. The lawsuit, which joins others led by authors, could greatly impact the future of AI development and content training practices. | | | |
| 4 | | IT services provider Accenture announced it's acquiring Udacity, the online learning platform known for its AI and tech courses. Financial details weren't disclosed. As part of the announcement, Accenture said it also launched LearnVantage, a tech training service in AI and data science for enterprises and governments. More: - Accenture plans to invest $1B in LearnVantage over three years.
- Once the deal closes, Udacity's 200+ employees will become part of the Accenture division overseeing LearnVantage.
- LearnVantage will also collaborate with Amazon Web Services, Microsoft, and Google Cloud to offer generative AI training and certifications to their joint customers.
- Kishore Durg, LearnVantage global lead, said generative AI's rapid growth has boosted client demand for cloud, data, and AI training and upskilling.
Zoom out: | | | |
| 5 | | A federal jury has charged an ex-Google engineer with stealing AI trade secrets for two Chinese companies he secretly worked for. Linwei Ding was arrested Wednesday and faces four counts of theft of trade secrets. More: - Ding is accused of stealing more than 500 secret files on Google's AI, mainly concerning its TPU chips used in AI operations.
- The files allegedly contained designs for Google's v4 and v6 TPU chips, specs for GPUs in its data centers, and machine learning workload designs.
- The government claims Ding began stealing files and was then offered a CTO position by the Chinese AI firm Rongshu.
- He later founded his AI startup Zhisuan, all while employed at Google.
- Dong stepped down from Google in December 2023 after the company asked about his file uploads.
- If convicted, he faces up to 10 years in prison and a $250,000 fine per charge.
| | | |
| 6 | | QUICK HITS - Top Performers are seeing a 56% increase in deal generation and productivity when they use this sales framework.*
- Microsoft is expected to unveil its first "AI PCs," the Surface Pro 10 and Surface Laptop 6, on March 21.
- Amazon Web Services launched Generative AI Competency, a program to help cloud customers validate the effectiveness of GenAI models, software, and hardware.
- Budget constraints at the National Institute of Standards and Technology (NIST) have led to infrastructure decay and reduced staffing, potentially undermining the agency's role in overseeing the next generation of AI models under the Biden administration.
* This is a sponsored post. | | | |
| Upcoming Events | | MAR
7 | | Portfolio Diversification - Learn from knowledgable experts on the latest trends in alternative investment assets like private equity, crypto, real estate, and the art market. | | | | | | | MAR
27 | | Discover how generative AI can speed up your enterprise, expand your business model, and improve your value proposition. Network with hundreds of AI leaders.* | | | | | | * This is a sponsored event | | | |
 | AI and technology writer | | Beth is a contributing editor and writer of Inside's AI and Tech newsletters. She has written for publications including USA Today, the Arizona Business Gazette, and The Arizona Republic, where she received recognition with a Pulitzer Prize nomination and a First Amendment Award for collaborative reporting on state pension cost increases. You can reach her at Beth.Duckett@yahoo.com. | | This newsletter was edited by Aaron Crutchfield | |
| |
 | RAD AI is the future of digital marketing. Execs from Hasbro & Skechers trust its tech to boost ROI. | |
|
| |
Comments
Post a Comment