| |
| Welcome to Inside AI! Here are today's top stories: - A study found OpenAI's language models have performed worse in some areas, including math problems and answering sensitive questions.
- A new Brookings report highlights the clustering of AI jobs across America.
- Google is developing an AI tool that writes news articles.
Beth
p/beth-duckett | |
| 1 | | A study by Stanford and Berkeley researchers has found that OpenAI's language models performed worse in some areas in June compared to March. For instance, GPT-4's accuracy in identifying prime numbers dropped from 97.6% to 2.4%. More: - The non-peer-reviewed study examined the performance of GPT-3.5 and GPT-4 in areas like solving math problems, answering dangerous/sensitive questions, generating code, and visual reasoning.
- GPT-4 showed less willingness to answer sensitive questions in June, and both models had more formatting mistakes in code generation.
- In June versus March, GPT-4's directly executable generations fell from 52% to 10%.
- The paper highlights the issue of model drift, or a decline in the models' accuracy and performance over time.
- "Overall, our findings show that the behavior of the 'same' LLM service can change substantially in a relatively short amount of time," the researchers wrote, adding that it's important to continuously monitor the models' performance.
Zoom out: - The study aligns with some user reports about the models becoming less intelligent.
- However, OpenAI's vice president of product, Peter Welinder, has denied intentional changes to make the models "dumber," saying that users may notice more issues over time simply because they use ChatGPT more.
|     | |
| 2 | | The news: According to a new study by the Brookings Institution, the rise of generative AI could further strengthen the San Francisco Bay Area's dominance in the technology industry. What it found: In May 2023, almost 60% of new generative AI jobs were found in the Bay Area and other early adopter metro areas like Seattle. Over the past 10 months, just six metro areas (San Francisco, San Jose, New York, Los Angeles, Boston, and Seattle) represented nearly half of U.S. generative AI job postings in the Brookings database. What it means: The report emphasizes the importance of making the development of the AI industry more geographically inclusive. Researchers also caution that the widespread adoption of generative AI may not automatically lead to widespread innovation, job growth, and leadership opportunities. To address this, it says major efforts should be made by policymakers and others to ensure that the benefits of AI are distributed more evenly across regions. 
Source: Brookings Institution | | | |
| |
|
| 3 | | Google is testing an AI tool that can generate news content, according to The New York Times. The tool, known internally as "Genesis," was showcased to executives from major media outlets including the Times, The Washington Post, and News Corp., owner of The Wall Street Journal. More: - According to the Times, Google has pitched the tool as "responsible" technology that could serve as a personal assistant for journalists, automating certain tasks to free up their time.
- The company said the AI tool could help with headline generation and strengthen writing styles.
- However, some executives expressed concerns about the tool overlooking the effort required for accurate and artful news stories.
Zoom out: - While some newsrooms have already started using AI-generated content, recent backlash against publications for publishing error-prone AI-generated stories highlights some of the current limits of AI in journalism.
| | | |
| 4 | | Elon Musk says Tesla plans to invest $1B in its Dojo supercomputer over the next year. The project, which was officially announced during Tesla's AI Day in 2021, aims to enhance Tesla's Full Self Driving (FSD) feature by training machine learning models on video data from its vehicles. More: - So far, Tesla has relied on an Nvidia-powered supercomputer, but it now aims for a custom-built solution and self-designed chips.
- This can provide increased control over confidential driver data following a recent data leak and limited availability of Nvidia's GPUs.
- Tesla has gathered over 300 million miles of video data from its vehicle cameras and sensors to enhance FSD through machine-learning training.
- The Dojo supercomputer, now expected to be completed by 2024, will leverage that data.
- Tesla aims to achieve an in-house compute capability of 100 exaFLOPS by next year.
| | | |
| A message from our sponsor, Inside.com. |  | | VC Investors Meet AI Founders at Exclusive Event Calling all AI Founders! Don't miss your golden opportunity to join the Two-Day VC Pitch Event, Meet Our Fund 5, in the vibrant landscape of the SF Bay Area. Engage with influential investors, including and hosted by Jason Calacanis, as they captivate and collaborate with visionary AI entrepreneurs like you. Immerse yourself in the dynamic world of venture capital, network with industry-leading professionals, and gain access to the resources and funding necessary to accelerate your startup's growth. Seize your chance and reserve your place among the 300 attendees today! Learn more | |
|
| 5 | | Nvidia is reportedly in discussions to make an equity investment in Lambda Labs, a startup that provides servers with its GPU chips. Lambda Labs is said to be raising $300M at a post-money valuation of $1B. More: - Since March, Nvidia has invested in seven private startups, primarily focusing on AI and cloud computing companies that also happen to be its customers.
- The chipmaker participated in Adept's $350M Series B, CoreWeave's $221B Series B, Cohere's $270M Series C, Synthesia's $90M Series C, Inflection's $1.3B funding round, Runway's $141M funding round, and Aleph Alpha's $212M investment since March.
- Some companies developing AI software are turning to Lambda, as well as rival Nvidia-backed CoreWeave, for computing capacity instead of relying on larger cloud providers.
- The demand for GPU servers has outpaced the supply from major providers.
For the latest on AI startup funding and more in the VC community, subscribe to our Inside Venture Capital newsletter. | | | |
| 6 | | Amazon is set to expand its biometric palm-scanning payment system, Amazon One, to all 500+ Whole Foods stores by the end of this year. With over 3 million users currently, Amazon One allows individuals to connect their Amazon accounts and payment cards to their palm prints. More: - Once a palm print is stored in Amazon's system, users can hover their hand over the scanner, and the system automatically deducts the payment from their linked Amazon account.
- Initially introduced in Amazon's Go stores in 2020, the palm scanner technology was later extended to over 200 Whole Foods locations, as well as various restaurants like Panera and travel retailers such as Hudson, CREWS, and OHM.
- In May, Amazon introduced an identification feature that enables users to purchase alcohol by scanning their hands.
- As Amazon stores users' biometric data in the cloud rather than on the device itself, there have been discussions about the potential privacy and security risks involved.
| | | |
| 7 | | QUICK HITS - Elevate your cloud expertise with the AWS Cookbook. Practical, concise, AWS-focused: This resource is your key to cloud success.*
- SAG-AFTRA, the prominent actors union, is on strike over concerns that AI and CGI advancements may curb the demand for human actors. Negotiations with studios and streaming services focus on protecting actors' rights and compensation for their likeness or voice being used.
- A report from the University of Oxford's Internet Institute found that AI companies employing gig workers for data labeling and other tasks consistently fail to meet basic labor rights standards. The work platforms included Amazon Mechanical Turk, Scale AI, and Appen.
- A U.S. district judge said he is leaning toward dismissing an artists' lawsuit against generative AI companies, including Stability AI and Midjourney. However, the judge said he would allow the artists to submit a new complaint with "more facts" about the purported copyright infringement.
- There are 12,000+ ways small businesses can qualify, but finding out if you're approved takes only two minutes. Apply now with no obligation, and discover what refund you're entitled to.*
*This is a sponsored post. | | | |
| Upcoming Events | | JUL
20 | | How to Set Up a Fundraising Campaign | | | | | | | JUL
25 | | Lacework hosts former CEO of Starbucks, Jim Donald, for a fireside chat on what CEOs need to know about cybersecurity.* | | | | | | | JUL
31 | | #TwoCents Discussion Community Challenge | | | | | | | AUG
2 | | Join Vanta and VioletX and find out how growing startups should navigate the complex compliance space.* | | | | | | | OCT
16 | | Meet Our Fund 5, Building and Investing in AI | | | | | | * This is a sponsored event | | | |
Term of the Day Lean startup: A lean startup occurs when founders and teams understand customer needs and refine their products accordingly to reduce wasted efforts and costs. Question of the Week Do you think it's ever acceptable to bring politics into the workplace? Join the conversation |
INSIDE AI LEADERBOARD (7 DAYS) |
 | AI and technology writer | | Beth is a writer and editor at Inside.com covering artificial intelligence and daily technology news. She has written the Inside AI newsletter since 2019. You can reach her at Beth.Duckett@inside.com. | | This newsletter was edited by Aaron Crutchfield | |
| |
| |
|
| |
|
| |
Comments
Post a Comment