| |
| Here are today's top stories: - Meta axed its ESMFold team, which developed a language model for predicting protein structures.
- A study found that AI text and image generators offer harmful advice on eating disorders.
- Zoom has clarified its AI training policy after a backlash over how it could use customer data.
Beth
p/beth-duckett | |
| 1 | | Meta, the parent company of Facebook, has disbanded its ESMFold team, which was focused on an AI project that predicts protein structures. As reported by The Financial Times, Meta is now prioritizing the development of AI products with profit potential, leading to a shift away from "blue-sky" scientific projects. More: - Meta employed about 12 scientists to work on ESMFold, which can predict the molecular structures of proteins from viruses, bacteria, and other microbes.
- The company also shared a database it created of more than 617 million of the "metagenomic" structures of proteins.
- This past spring, the ESMFold team was disbanded as part of larger layoffs within the company.
Zoom out: - Meta's focus is now on developing generative AI products, including chatbots.
- Since February, Meta's product chief Chris Cox has led a new generative AI team with "several hundred" employees.
- The company is now gearing up to launch AI-powered chatbots with distinct personalities, called "personas," on Facebook and Instagram as soon as next month.
|     | |
| 2 | | The news: A report by the Center for Countering Digital Hate (CCDH) reveals that AI tools, including ChatGPT and Bard, have a tendency to provide harmful advice and content related to eating disorders. Details: The study involved tests on popular AI tools, including ChatGPT, Bard, My AI, DreamStudio, Dall-E, and Midjourney, using 20 prompts associated with eating disorders, such as "thinspo" diets and vomiting-inducing drugs. The tools suggested harmful behaviors like hiding food and planning low-calorie diets. What the numbers say: Before the use of "jailbreaks" to bypass safety restrictions, Snapchat's My AI refused to generate harmful advice and encouraged seeking help from medical professionals. However, after jailbreaks, ChatGPT generated harmful responses to all 20 prompts, Bard to eight, and Snapchat's tool to 12. Overall, the study found that the AI tools generated harmful text and images 41% of the time. Why it matters: Psychologists are warning that AI-generated content can be particularly dangerous for individuals with eating disorders, potentially exacerbating mental health issues. The CCDH says tech companies aren't doing enough to prevent AI tools from disseminating harmful content and recommends they design products to be safer and test them "rigorously" before launching them to the public. 
| | | |
| A message from our sponsor, Revelry. |  | | Revelry ships gold every week, one brick at a time. Building software can be hard. This is why Revelry optimized product development workflows into a process it calls Lean Agile. The company has also built an arsenal of tools to de-risk various delivery tasks. The combination of Revelry’s process and stack allows for incredible efficiency - helping engineering teams ship gold every week, one brick at a time. According to Revelry, its Lean Agile approach positions product teams to form and deliver 30-40% faster than industry averages. To learn more, download its Lean Agile guide. Download Now | |
|
| 3 | | Zoom has clarified that it will not train AI models on customers' audio and video data without their consent. The clarification comes after Zoom updated its terms of service, stating that it plans to use certain customer data, like product usage, telemetry, and diagnostic information, for AI training. More: - The new terms, effective from July 27, allow Zoom to use "service-generated data" for AI training, but the company assures that audio, video, and chat content won't be used without customer consent.
- Users who turn on Zoom's generative AI features will be asked to provide consent for their individual customer content to be used for improving AI services.
- A toggle button lets users switch Zoom's access to the data on or off.
Zoom out: - Zoom recently introduced two new AI features, Zoom IQ Meeting Summary and Zoom IQ Team Chat Compose, which summarize meetings and compose chats, respectively.
| | | |
| 4 | | British researchers trained a deep learning model to predict a person's computer keystrokes based on audio with up to 95% accuracy. The acoustic attack could pose a threat to data security by leaking passwords and other sensitive information. More: - The "CoAtNet" classifier detects minute differences in keystroke sounds to make predictions, relying on audio recorded from a nearby microphone or via a Zoom call.
- To test the algorithm, the team used a MacBook Pro and iPhone. According to the research, the classifier achieved an impressive 95% accuracy when trained on keystrokes recorded by a nearby phone.
- Even when using Zoom, the accuracy only dropped to 93% and 91.7% for Skype.
- The paper discusses ways to safeguard users against these types of attacks, including altering typing styles, using randomized passwords, and turning on biometric authentication or password managers.
| | | |
| A message from our sponsor, Airplane. |  | | Airplane is the developer platform for building custom internal tools. You can transform scripts, queries, APIs, and more into powerful UIs and workflows in minutes. Use Airplane to build internal admin panels, host scheduled operations, share scripts and on-call runbooks, create complex multi-step workflows, and more. Airplane is code-based: everything you create in Airplane can be version controlled, integrated with the rest of your codebase, and extended using third-party libraries. We have Tasks (single or multi-step functions), Views (React-based platform to create custom UIs and dashboards using our component and template library), and offer enterprise-grade defaults including maintenance-free schedules, permissions and RBAC, audit logs, notifications, self-hosting, and approval flows. Try for free! | |
|
| 5 | | Palantir Technologies raised its annual revenue forecast amid "unprecedented" demand for its AI platform. The Artificial Intelligence Platform, AIP, has gained traction with over 100 organizations, and the company is in talks with 300 more enterprises. More: - AIP is designed to improve military and enterprise decision-making through large language models.
- The platform is designed to incorporate LLMs, like OpenAI's GPT-4 or Google's BERT, into privately-operated networks.
- Palantir said it can be used by militaries to aid in battlefield decision-making and intelligence by tapping into AI models.
- A video demo shows an AI-powered chatbot used for tasks like drone reconnaissance, attack planning, and disrupting enemy communications.
Zoom out: - Since it launched in April, shares of Palantir have more than doubled.
- The company is now aiming for S&P 500 inclusion as it approaches the year milestone, with three consecutive profitable quarters.
| | | |
| 6 | | AI Events and Conferences: - Ai4 2023: Aug. 7-9, Las Vegas.
- Efficient Generative AI Summit: Sept. 11, Santa Clara, Calif.
- AI Hardware Summit: Sept. 12-14, Santa Clara, Calif.
- Edge AI Summit: Sept. 12-14, Santa Clara, Calif.
- MLCon NYC 2023: Sept. 25-28, New York, N.Y., or online.
- AI & Big Data Expo: Sept. 26-27, Amsterdam, The Netherlands.
- AIAI Computer Vision Summit — Berlin: Oct. 5, Berlin, Germany.
- WorId Summit AI: Oct 11-12, Amsterdam, The Netherlands.
- AI World Barcelona: Generative AI Conference: Nov. 14-15, Barcelona, Spain.
- AI & Big Data Expo Global 2023: Nov. 30-Dec. 1, London, U.K.
- The AI Summit New York: Dec. 6-7, New York, N.Y.
| | | |
| 7 | | QUICK HITS - Wiz partnered with Wiley to create the AWS Security Foundations Guide For Dummies. This free 46-page guide includes techniques for securing AWS resources.*
- SEC Chairman Gary Gensler expressed concerns that AI could lead to future financial crises due to "herding," when all U.S. business systems rely on the same information generated by only two or three foundational models.
- Google has added an AI-powered grammar checker to Search, competing with Grammarly.
- Microsoft's Bing Chat AI chatbot is now available on third-party browsers like Safari and Chrome, in addition to its Edge mobile browser and standalone Android and iOS apps.
- The Dungeons & Dragons franchise has banned the use of AI technology for creating artwork and updated its guidelines after a controversy over an AI-generated illustration.
- Join over 5,000 fast-growing companies around the world that trust Vanta to manage risk and prove security in real time. Inside readers get $1,000 off.*
*This is a sponsored post. | | | |
| Upcoming Events | | AUG
5 | | Learn about streamlining SOC 2 compliance with Vanta and AWS. Join the webinar on Aug 17.* | | | | | | | OCT
4 | | Join this free hybrid event in Amsterdam and you'll hear from industry experts at Meta, Opera, AdGuard, Mozilla, Google, and more.* | | | | | | | OCT
16 | | Meet Our Fund 5, Building and Investing in AI | | | | | | | NOV
7 | | Join 300+ CCOs and senior leaders from Amex, Verizon, Hilton, Wells Fargo and more at Reuters Events: Customer Service & Experience East 2023.* | | | | | | * This is a sponsored event | | | |
Term of the Day Activation: Marketing activation is the process of reaching a specific target audience and increase brand or product awareness and engagement. Question of the Week Do you think the 9-to-5 workday still relevant in today's modern workplace? Join the conversation |
INSIDE AI LEADERBOARD (7 DAYS) |
 | AI and technology writer | | Beth is the curator and writer of the AI and Tech newsletters at Inside. With a career in journalism and editing that began in 2007, she has written for publications including USA Today, The Arizona Republic, Arizona Business Gazette, and more. You can reach her at Beth.Duckett@yahoo.com. | | This newsletter was edited by Aaron Crutchfield | |
| |
| |
|
 | Build custom internal UIs and workflows quickly using code with Airplane. | |
|
| |
Comments
Post a Comment