Alibaba’s new AI model beats GPT-4o on benchmarks

Hello, AI Enthusiasts!

Welcome to FavTutor’s AI Recap! We’ve gathered all the latest and important AI developments from the past week in one place, just for you.

In Today’s Newsletter: 😀 

• Alibaba’s new AI model beats GPT-4o on Benchmarks

• Major News Websites Blocking Apple’s AI Crawlers

• Magic's new AI model can process 10 million lines of code

• OpenAI & Anthropic signed deals with the US government

• Google restricting its AI products before US elections

Alibaba
🏅 Alibaba’s new AI model beats GPT-4o on Benchmarks

Alibaba’s latest Qwen2-VL has shown better video understanding, function calling and multilingual text-image processing. It can analyze handwriting in multiple languages including Japanese, Korean, Arabic, and Vietnamese.

Insights for you:

• Alibaba Cloud has launched its new Vision Model called Qwen2-VL.

• On many benchmarks, it has surpassed many big LLMs like GPT-4o, Claude 3 Haiku, and Gemini 1.5.

• The model can analyze high-quality 20+ minutes long videos and answer questions about the content of the video in real time.

Together With Guidde
That meeting could have been a Guidde

Manifest this: instead of summoning the entire customer team for another round of snooze-inducing meetings, why not let Guidde work its AI magic? Transform your instructions into professional looking how-to videos that'll have your team actually engaged.

You might be wondering, but how? This is how:

• Guidde automatically creates a storyline with highlights

• You can add & translates text to voice in over 40+ languages

• Record once, generate a video + pdf + mp4 + html… for SEO value

• Integrate with Zendesk, Notion, Slack, Salesforce and more.

Simply click capture on the browser extension and the AI will automatically generate step-by-step video guides, complete with visuals, voiceover, and call to actions.

The best part? You can try Guidde for free

Apple
👎️ Major News Websites Blocking Apple’s AI Crawlers

According to a new report by WIRED, major news sites like The New York Times, The Financial Times, The Atlantic etc. have opted out of using their content to train Apple’s AI models. However, some of these websites have done licensing deals with AI companies to properly get compensation for helping the AI training. This might be a way to pressure AI companies for licensing.

Insights for you:

• The following websites are blocking their data for Apple’s AI: The New York Times, The Financial Times, The Atlantic, Vox Media, and the USA Today network.

• They are using the ‘Applebot-Extended’ tool to opt out of the AI training, but still, the content is accessible for Apple’s general search products like Siri.

• In a similar study of 1,000 websites by Originality AI, about 7% of websites were found to be blocking Applebot-Extended.

Magic
🔮 Magic's new AI model can process 10 million lines of code

Magic AI has unveiled an LTM-2-mini model that is making records in the AI domain. It can process 100 million tokens of context, which is equivalent to about 10 million lines of code. Their focus is on software development, as with such a massive context window, it can access the project's entire code and libraries.

Insights for you:

• Magic AI’s new language model: LTM-2-mini; can work with a 100 million tokens context window.

• The company has also partnered with Google Cloud to build supercomputers powered by Nvidia’s GPUs.

• It has recently raised $320 million from a group of investors last week.

Anthropic & OpenAI
🏛️ OpenAI & Anthropic signed deals with the US government

According to the official agreement, the U.S. Artificial Intelligence Safety Institute announced formal collaboration on AI safety research, testing and evaluation with both Anthropic and OpenAI. These agreements come at a time when AI companies are facing increasing scrutiny from regulators.

Insights for you:

• This is a first-of-a-kind deal between AI companies and the government.

• The US AI Safety Institute will now have access to upcoming AI models before public release.

• The aim is to evaluate safety risks and provide feedback to OpenAI and Anthropics to put safeguards on them.

Google
😶 Google restricting its AI products before US elections

To bring more transparency to the US elections, Google announced several safeguards in the latest blog post. This is because spreading information in the election is a crime and AI tools are also a part of the system now. AI tools are known to produce incorrect information and this might lead to a big problem for the AI companies.

Insights for you:

• Google is restricting the use of its AI products for the upcoming 2024 U.S. elections, including Search AI overviews and YouTube AI summaries for live chat.

• They would also not answer election-related queries on the Gemini apps and web experience.

• This also includes embedding watermarks when you create an image using Gemini’s Imagen 3 model.