Thread by @whoisvaibhav

A mega tweet thread for rounding up various AI-related events (no way to cover them all) that have happened this year so far.

#AIisComing #AIisHere
(1/many)

The @verge reported earlier this month that Microsoft is looking at ways to integrate @OpenAI's GPT language model in products such as Word, Outlook, etc.

There are already some AI-powered capabilities built-into Word through the Editor feature.
(2/n).

theverge.com/2023/1/9/23546…

It helps with grammar, tone (casual, formal, professional), checks for plagiarism, auto-suggests text, and can even create summaries.

(3/n)

Microsoft unveiled a text-to-speech (TTS) model called VALL-E that can generate a clone of anyone's voice after listening to a 3-seconds recording.

TTS is going to become increasingly common. Companies such as @elevenlabsio already offer paid services in this area.

(4/n)

Here's the github demo page for VALL-E: valle-demo.github.io

This type of technology deep potential for disability assistance, education, customer service, entertainment and much more.

(5/n)

Rewriting reality in real-time is going to get quite common. @nvidia released an eye-contact feature that helps you make eye contact on video calls. It does so by painting your eyes in real-time over your video feed (see my picture):

(6/n)
arstechnica.com/information-te…

But this one isn't that new. Both Apple (for its iOS devices) and Microsoft (for Surface X devices) have offered this for some years:

(7/n)
blogs.windows.com/devices/2020/0…

Some of the implementations of these features are creepy to say the least. But these will only get better from here and become the norm.

(8/n)

#ChatGPT has surpassed Instagram in the daily active users and is growing faster than TikTok.

It has been able to clear advanced human examinations such as CPA, the bar, and medical board exams.

Schools have started banning access.

(9/n)

Amidst all this, OpenAI is starting to monetize ChatGPT through a Plus/Pro plan.

(10/n)

techcrunch.com/2023/02/01/ope…

BTW, there is so much happening that DALL-E 2 and Imagen from last year are already old news. Back to the thread. I still have a lot to cover.

(11/n)

But talking about Text-To-Image, Google released Muse, a Text-To-Image model that is significantly faster than Dall-E 2 or Imagen.

(12/n)

muse-model.github.io

For that matter, @StabilityAI announced @deepfloydai and teased a text-to-image model called IF which handles fonts and text in images beautifully. Check out their Twitter handle for samples of what it can do.

(13/n)

As generative AI becomes more ubiquitous, the need to detect AI-generated is being felt. @edward_the6 launched #GPTZero. @elevenlabsio committed to releasing a tool to detect the use of their AI.

(14/n)

https://twitter.com/elevenlabsio/status/1620443168851525634?s=20&t=xTv34dHINqBefOg7KpSd8A

Going back to TTS - @Apple launched an audiobook program that narrates books using AI-generated voices.

As TTS mimics human voice, intonation, and emotion better, this will become very standard.

(15/n)

authors.apple.com/support/4519-d…

ex-OpenAI folks raised hundred of millions, founded @AnthropicAI and are beta testing a #ChatGPT competitor (called Claude) that is supposed to be better.

(16/n)

techcrunch.com/2023/01/09/ant…

With Sundar Pichai declaring a code red, we may even see Alphabet's #ChatGPT competitor released this year - @DeepMind's Sparrow.

(17/n)

deepmind.com/blog/building-…

Another one of @DeepMind's AI's, AlphaFold, was credited for a first-of-its-kind drug discovery for human application. This kind of application will continue to have huge benefits in the field of medicine.

(18/n)

phys.org/news/2023-01-a…

A lot of these AI capabilities aren't just fun and games. Generative AI use cases are rapidly coming to real-world uses. @Scenario_gg creates game assets using generative AI. They just raised $6 million.

(19/n)

techcrunch.com/2023/01/19/sce…

Another startup that raised money is @character_ai which allows you to create a character with a personality and then chat with it about anything you want. Below is a screenshot of me chatting with a Tony Stark character someone made.

(20/n)

Dall-E most directly threatened the livelihood of stock photographers. Well, @Shutterstock, a stock photo company, launched a generative-AI image generation service that allows you to make your own images and license them.

(21/n)

shutterstock.com/generate

But not just stock photographers, authors too, @CNET was found using AI to publish entire articles on their website:

(22/n)

futurism.com/the-byte/cnet-…

And @BuzzFeed was more upfront with their plans regarding AI in content creation with an announcement of OpenAI partnership and plans to use ChatGPT.

(23/n)

dataconomy.com/2023/01/buzzfe…

Text-to-music is another area that may become a thing. Google released a paper for a model called MusicLM that makes "high-fidelity music from text descriptions". Check it out at the link below.

(24/n)

google-research.github.io/seanet/musiclm…

Well, I have run out of time, so the thread ends here. If you enjoyed the thread, consider liking, commenting, and/or retweeting.

(n/n).

@threadreaderapp unroll