AI Singularity.com丨AI News丨AI Tool Recommendation丨AI Tool Tutorial
Provide 1000+ AI efficiency tools丨 AIEVA.CN 丨Welcome to visit
AI Singularity reported on June 21, 2024丨AI Information Morning Post
Anthropic发布新一代Claude 3.5 Sonnet，测评超越GPT-4o
On June 20, local time, Anthropic, one of OpenAI's strongest competitors, released the most powerful AI model to date, Claude 3.5 Sonnet. In several performance tests, the Claude 3.5 Sonnet outperformed competitor AI models such as GPT-4o. Subsequently, the company will also release other large models of the Claude 3.5 series.
Anthropic 推出最新模型 Claude 3.5 Sonnet 宣称该模型在各种任务上的表现与 GPT-4o 和 Google Gemini 处于同一水平甚至更好。
The Sonnet model is Anthropic's mid-range "Big Cup" model, which sits between the small Haiku model and the highest-end model Opus, and according to Anthropic officials, the Claude 3.5 Sonnet is even stronger than the previous generation Claude 3 Opus.
According to Anthropic, Claude 3.5 Sonnet will perform better at writing and translating code, handling multi-step workflows, interpreting diagrams, and transcribing text from images, as well as better at understanding humor and being able to write in a more human way.
In seven of the nine benchmarks, and four of the five visual benchmark models, the Claude 3.5 Sonnet outperformed GPT-4o, Gemini 1.5 Pro, and Meta's Llama 3 400B.
In addition to the new models, Anthropic has also introduced a new feature called "Artifacts", which expands the way large models interact with users. Allow users to integrate AI applications directly into workflows. For example, if you ask Claude to write an email, you can edit the email directly within the Claude app without having to copy it into a text editor.
WeChat input method released version 1.2 upgrade, which can directly conduct AI Q&A
On June 20th, WeChat officially released a new version of the WeChat input method, launching the 1.2.0 official version update, which has been launched on Windows and macOS platforms. The main change in this update is the embedded AI Q&A function in the software, which allows users to enter the content information of the question and click the "=" equals button to directly get the AI answer.
The AI Q&A function can not only provide detailed introductions, but also give corresponding information feedback based on the keywords entered by the user.
The upgrade log shows that, in addition to AI Q&A, the WeChat input method also supports enable/disable emoji and emoji recommendations in the settings panel; Support time and date input in more formats.
Foreign media exposed: "Apple AI" is seeking cooperation with Chinese companies and has contacted Baidu Ali Baichuan
According to foreign media reports, Apple seeks to cooperate with local AI model developers in China, including Baidu and Alibaba, as well as AI startup Baichuan Intelligence, aiming to provide its "Apple Intelligence" service in the Chinese market to meet local competitors in the Chinese market who integrate artificial intelligence into mobile phones.
At the WWDC developer conference held in June this year, Apple released an important product demonstration in the field of artificial intelligence, "Apple AI", showing AI tools such as custom emoji generators to iPhone users in the United States. However, the company has not said anything about how to promote AI services in China, the second largest consumer market for iPhones.
According to market research firm Counterpoint Research, the iPhone's share of shipments in the Chinese market has dropped to third, surpassed by two major local brands. Apple has previously considered introducing large language models into Chinese devices, but has not yet made substantial progress.
At present, the most powerful contender is not intended to be Baidu's Wenxin Yiyan, but Ali's strength should not be underestimated. Luca Maestri, Apple's CFO, said in an interview last month: "China is the most competitive market in the world, but we are confident in maintaining our market position!" ”
"Apple AI" can only be provided to flagship models? Apple executives responded
Apple's WWDC conference unveiled the "Apple Intelligence" service to showcase its powerful AI capabilities. However, it is limited to the iPhone 15 Pro, iPhone 15 Pro Max, and iPad and Mac devices with the M1 chip or later.
Apple explained that this is because the inference and computing requirements of large language models are extremely high, and if Apple deliberately wants to "force" users to upgrade and buy the latest models, it can limit the iPad and Mac series products to only be equipped with "Apple AI" on the latest models.
Ming-Chi Kuo, a well-known Apple analyst, believes that whether "Apple AI" can provide carrying services depends on the size of the DRAM of the device itself, rather than the AI computing power. For example, the A16 chip equipped with the iPhone 14 has the computing power to meet the requirements of running Apple's 3 billion device-side models, but the DRAM is only 4G, which is not enough to run large models.
Well-known foreign media accused the AI search tool Perplexity of scraping its own content without permission
Through investigation and analysis, Wired magazine found that the AI tool Perplexity seems to ignore the "bot exclusion protocol" and quietly scrape content from some websites that don't want to be visited by bots, including Wired magazine itself.
Wired magazine said it had blocked Perplexity AI's access permission in robots.txt files, the standard documentation for web crawlers, but Perplexity AI was still trying to access their content. They found that the text generated by Perplexity AI under relevant news headlines was highly similar to news articles written by Wired.
The outlet also set up a new website to monitor server logs, then prompted a bot in Perplexity AI to summarize the site's content, and the site's server immediately recorded a suspicious secret IP address visiting the site.
Wired isn't the only well-known outlet that restricts Perplexity AI from scraping its own content, as Forbes, a leading financial magazine, questioned Perplexity's response as a direct copy of its own content.
Beijing's first "AI face swapping software" infringement case was announced: the photo of a national style blogger was stolen and became a commercial template
On June 20, according to the news released by the Beijing Internet Court, the court opened a first-instance trial on the same day and pronounced judgments in two infringement cases of "AI face swapping software" in Beijing, finding that the defendant used the screen of other people's videos to make a general commercial template after the "AI face swapping" operation, and then provided a network service of "face swapping" service, and the provider of the software platform infringed on the plaintiff's personal information rights and interests.
The plaintiffs in the two cases, Liao and Wu, are both Chinese-style short video models and have many fans on the whole network. The defendant was the operator of an "AI face-swapping" app.
The plaintiff asserted that without its authorization and consent, the defendant used the plaintiff's appearance video to make a face swap template, uploaded it to the face swap app in question, and provided users with paid use for profit. The plaintiff argued that the defendant's actions infringed the plaintiff's portrait rights.
After the trial, the court held that the defendant's actions constituted an infringement of the plaintiff's rights and interests in personal information. The court of first instance ruled that the defendant apologized to the plaintiff and compensated for moral and economic losses. The case is still in the appeal cycle.
[This article is from丨AI Singularity Network丨The same name of the whole network account丨Welcome to search and follow]
AI Singularity丨Provide 1000+ AI efficiency tools丨 AIEVA.CN 丨Welcome to visit