In this week, Apple WWDC, known as the "Technology Circle Spring Festival Gala", was held, and "machine learning" became a term that was mentioned many times that night. In the same week, the college entrance examination came, but this year, in addition to focusing on the third-year high school candidates who went to the examination room, the popular AI has also become an alternative focus of "new students rushing to the exam".

Apple holds WWDC23 developer conference

Many references to "machine learning"

In the middle of the night of June 5th, Apple held a WWDC23 keynote speech. Of course, the focus was undoubtedly the surprise release of the $3,499 MR device Vision Pro.

But throughout the event, the words "Machine Learning" and "Transformer models" carried through the keynote speeches throughout the night.

Hardware: serving usage scenarios

In terms of hardware, Apple released the M2 Ultra chip with a maximum of 192GB of unified memory. It also emphasized that "using such a large unified memory can run other PCs that cannot undertake machine learning operations, such as using it to run Transformer large models". The intention is obvious.

At the application level, AirPods Pro can also "learn" to automatically adjust the volume through machine learning, and fine-tune the volume experience according to the user's preferences at different time periods and the surrounding environment.

Software: Focus on on-device machine learning

In terms of software, iOS 17 is also enhanced by "machine learning". First on "Call Live Voicemail" and iMessage, iPhone will be able to provide local instant voice transcription content.

The "automatic correction" of keyboard input, with the help of the powerful computing power of the Apple chip, will be able to "every time a key is pressed, the iPhone will run the Transformer language model once", and the local prediction and learning capabilities perform well.

The newly added Notes APP can also mark interesting moments in the user's life through local machine learning, create personalized moment suggestions, and discover a new self when the user re-views them.

From this keynote speech, we can really understand Apple's approach to "machine learning": implement all privacy-related models to run locally on the device, and pursue the efficiency and practicality of machine learning.

This also shows Apple's clear attitude towards the development of machine learning and even AI: these technologies should serve human beings and respect privacy.

The development of Midjourney v6 is currently blocked

The expected release time has been delayed

On June 7th, Midjourney held an "Office Hours" user exchange meeting on the official Discord channel, answered users' questions, and introduced their current situation.

In this exchange meeting, the Midjourney development team admitted that the current development of v6 is behind expectations, and they must refactor the new version and conduct a new round of training, which requires them to clean up the data, find better data and organize it , although the current refactoring is progressing smoothly, it is expected that the release date will be delayed by another month than expected.

The upcoming v5.2 version, although less updated, may have a good performance on "outpainting" (extended repainting).

The team is currently prioritizing building a website and mobile page independent of Discord, which can also be seen in their ambition to extend the service to a wider audience.

Bing chat will soon support third-party browsers

The function is being opened to some user experiments

Over the weekend, Reddit users showed screenshots of Bing Chat running in Safari and Chrome browsers.

Later, foreign media confirmed to Mikhail Parakhin, head of Microsoft's advertising and network services department, that Microsoft is trying to make Bing chat support third-party browsers other than Edge. It is currently conducting experiments and will gradually roll out to more users.

Chinese students build ControlNet for QR Code

Use SD to make the QR code "beautiful"

Recently, a group of very beautiful "QR codes" has aroused heated discussions among Reddit netizens, saying that it is a "QR code" rather than a "2D painting". . People exclaimed: "It can really be swept out!"

Such a "two-dimensional painting" is called ControlNet for QR Code by the author, which is the kind of "ControlNet" familiar to Stable Diffusion users. This also means that as long as some appropriate operations are performed, people can use AI to generate as beautiful The "two-dimensional painting".

It was Ni Hao, Chen Baiyu, Wang Zhaohan, and Chen Zhiyong who created the "two-dimensional painting". After a long period of ControlNet training, the four students used a combination of Checkpoint + LoRA + QR Code ControlNet to create these beautiful recognizable QR codes.

At present, they have uploaded some LoRA patterns to Civitai. The generation model of ControlNet for QR Code is still being optimized. Development updates, model releases, and technical documents will be announced in the future. Interested friends can look forward to it.

Text Generating Video Tool Gen-2 Starts Free Trial

The era of "Vensheng Video" has come

Runway announced that the video generation model Gen-2 is open to all users. Users can directly use prompt text/prompt images to create and edit video content, and support browsers and mobile APPs.

For example, input "rough waters" and Gen-2 can generate a short video of the waves, which is pretty good.

Gen-2 free users currently have a video generation quota of 105 seconds, and support paid upgrades to video duration, resolution, and generation speed.

Hot discussion on the Internet about "AI rush exam"

See what kind of college entrance examination composition AI writes

Every year's college entrance examination composition is the focus of discussion on the entire network, but this year's protagonist is not only senior high school candidates, but also AI. With the popularity of AI in recent months, people are also curious about what kind of answers these AIs at home and abroad can hand in in the college entrance examination composition.

Our Haoshuo AI also came to show off its writing skills. At the end of the article, let's take a look at what kind of answers it will hand over the composition questions of the National Paper A.

Question content:

Read the material below and write as required. (60 points)

People have better control over time due to technological development, but some people have become servants of time.

What kind of association and thinking did this sentence arouse in you? Please write an article.

Requirements: Select the right angle, determine the concept, clarify the style, and draw up the title by yourself; do not copy, do not plagiarize; do not disclose personal information; no less than 800 words.