- Nex AI News
- Posts
- Why Apple is taking a small-model approach to generative AI
Why Apple is taking a small-model approach to generative AI
Wednesday - June 12 - 2024
Among the biggest questions surrounding models like ChatGPT, Gemini, and Midjourney is their role in daily life.
Apple aims to answer this with Apple Intelligence, unveiled at WWDC 2024.
The company emphasized the capability of its in-house models during the presentation.
When generative AI tools like ChatGPT launched in late 2022, the finance industry quickly saw their potential to speed up data gathering and research.
Now, startup Linq has entered this space with an AI agent that automates financial analysis and research tasks.
š» NVDA NVIDIA Corporation | $120.91 | -0.87 |
š AAPL Apple Inc. | $207.15 | +14.03 |
š GME GameStop Corp. | $30.49 | +5.66 |
š» NIO NIO Inc. | $4.4600 | -0.2600 |
š» TSLA Tesla, Inc. | $170.66 | -3.13 |
Todayās newsletter :
Why Apple is taking a small-model approach to generative AI
Linq raises $6.6M to use AI to make research easier for financial analysts
FTC Chair Lina Khan shares how the agency is looking at AI
Researchers use large language models to help robots navigate
The Tech Worldās Greatest Living Novelist Goes Meta
Danish Media Threatens to Sue OpenAI
AI Is Appleās Best Shot at Getting You to Upgrade Your iPhone
Tech Lobbying Group Launches Campaign for AI Fair Use Rights
Singapore Lands $7.8B Semiconductor Plant for Automotive, Industrial Chips
Among the biggest questions surrounding models like ChatGPT, Gemini, and Midjourney is their role in daily life. Apple aims to answer this with Apple Intelligence, unveiled at WWDC 2024. The company emphasized the capability of its in-house models during the presentation.
Appleās approach to generative AI is pragmatic, focusing on user experience with smaller, customized models designed for its operating systems.
Apple Intelligence seamlessly integrates into the OS, prioritizing transparency and privacy.
Third-party models like OpenAIās ChatGPT will be utilized for broader information needs, with users opting in to share data externally.
Private Cloud Compute ensures privacy standards are maintained across devices and servers.
Appleās bespoke models are trained on licensed datasets and public data, with publishers having the option to opt out of AI training.
A whitepaper outlines Apple's principles: empowering users, representing diverse users, careful design, and protecting privacy.
Balancing user experience with transparency and privacy is crucial as the betas become generally available.
When generative AI tools like ChatGPT launched in late 2022, the finance industry quickly saw their potential to speed up data gathering and research.
Now, startup Linq has entered this space with an AI agent that automates financial analysis and research tasks.
Linq raised $6.6 million in funding led by InterVest and Atinum, with TechStars, Kakao Ventures, Smilegate Investment, and Yellowdog also participating.
Founded by MIT alumni Jacob Chanyeol Choi and Subeen Pang, Linq emerged after winning the Samsung Open Collaboration in 2023.
Linq's AI uses domain-specific search and large language models to automate tasks such as scheduling, communication, scanning research reports, and building financial models.
It can summarize securities filings, earnings reports, and call transcripts.
Linq plans to offer B2C tools for AI equity research, supporting portfolio managers with data synthesis.
Competing with Bloomberg, S&P, and startups like Fintool, Linq aims to stand out by offering an end-to-end service managing workflows and providing access to a wide range of data, including live transcriptions in multiple languages.
With customers like Samsung Financial Network and KPMG US, Linq will use its new capital for product development, hiring, and expansion into the Americas, Asia, and the Middle East.
The two-year-old startup has 12 staff and began generating revenue last October.
The U.S. Federal Trade Commission (FTC) will examine the rise of AI technology across all fronts, said FTC Chair Lina Khan at TechCrunchās StrictlyVC event.
Khan emphasized the agencyās goal is not to stifle startups with increased regulation but to ensure fair competition.
Khan noted an increase in consumer complaints, such as voice-cloning fraud.
The FTC is monitoring all AI aspects, from chips to apps, and seeking input on opportunities and risks.
The FTC faces challenges in policing AI, despite hiring technologists to help.
With fewer staff than in the 1980s, the agency uses innovative tactics like voice-cloning challenges to fight AI fraud.
They also focus on true openness in AI and policing AI hype to prevent deceptive advertising.
Someday, you may want your home robot to carry laundry downstairs and load it in the washing machine.
For an AI agent, this task is complex. Current methods use multiple hand-crafted machine-learning models requiring extensive visual data.
Researchers from MIT and the MIT-IBM Watson AI Lab developed a navigation method using language-based representations instead of visual data.
Their method converts visual observations into text captions, processed by a large language model to determine the robot's actions.
This approach allows for efficient generation of synthetic training data, though it doesn't outperform vision-based techniques.
Combining language-based and visual signals improves navigation performance.
This method, which uses less computational power and generates human-understandable trajectories, offers advantages in environments lacking visual data.
It also simplifies adapting to varied tasks and environments. However, it loses some depth information compared to vision-based models.
Funded by the MIT-IBM Watson AI Lab, this research aims to enhance AI navigation using language.
News
What else is nex?
The Tech Worldās Greatest Living Novelist Goes Meta
During my extended hang with Robin Sloan, tech worldās greatest living novelist, he shares a profound insight about science fiction.
I glance at my voice recorder: āMemory is full!ā I donāt panic but mentally spiral.
Danish Media Threatens to Sue OpenAI
In a new clash between AI and media, major Danish newspapers and TV stations are threatening to sue OpenAI unless compensated for allegedly using their content to train its models.
Karen RĆønde, CEO of the Danish Press Publicationsā Collective Management Organization (DPCMO), representing 99% of Danish media outlets, demands remuneration for their work.
AI Is Appleās Best Shot at Getting You to Upgrade Your iPhone
APPLEāS NEW AI strategy may drive its upgrade-your-iPhone push. At its annual developer conference, Apple announced Apple Intelligence, a suite of new AI features trained on massive datasets.
These features, integrated into existing apps, include turning sketches into images, sorting photos, rewriting emails, and enhancing Siri.
However, they will only work on newer devices like the iPhone 15 Pro, iPads with the M1 chip, and newer Macs.
The technology lobbying group Chamber of Progress has launched the "Generate and Create" campaign to support fair use protections for generative AI training and output.
The campaign highlights the creative benefits of generative AI and defends companies like OpenAI's right to scrape the internet for data to train their models, a practice facing increasing scrutiny from copyright holders.
Singapore Lands $7.8B Semiconductor Plant for Automotive, Industrial Chips
Singapore will host a new $7.8 billion semiconductor plant for the automotive, industrial, and mobile markets.
This joint venture between Vanguard International Semiconductor Corporation (VIS) and NXP Semiconductors will form VisionPower Semiconductor Manufacturing Company.
Reply