AIcceleration

Generative AI: 1 month of Incredible Announcements or an AIcceleration for the future of business

Aside the phenomenal launch of ChaGPT and ChatGPT4 (100 million users within 2 months of opening), Generative AI will impact numerous industries and jobs, from advertising/marketing to project management and HR. Goldman Sachs even predicts that AI could replace 300 million jobs... But I think Generative AI will create numerous new one too. There will definitely be a massive change for the future of work and productivity around AI augmented workers.

As part of our new training/keynote on generative AI (?? request a keynote or workshop for your executives), in collaboration with our experts at the HUB Institute, here is a summary of the many key announcements from the last 21 days!

After is successful launch, ChatGPT has announced ChatGPT Plugins to allow ChatGPT to connect via API with your brand application to better inform the user directly from ChatGPT (while potentially losing traffic and ad revenue).

In addition, with a new Retrieval plugin, ChaGPT will, for example, keep a memory of conversations to perform certain tasks. And this is without counting on ChatGPT5, already in development ??.

Researchers at Stanford have developed Alpaca AI, a local version close to ChatGPT 3.5 (based on Meta's LLaMA solution whose source code leaked publicly) with a training cost of $600. Pretty crazy to run such a solution on a simple laptop. A usage that violates Facebook's license agreement, which has since been stopped.

The intelligent assistant Anthropic Claude was launched in collaboration with Notion, Quora, and DuckDuckGo.

But Microsoft has also announced the upcoming arrival of CoPilot in the Office suite. You can now ask the generative AI integrated into Word or PowerPoint to assist you in pre-writing a note, translating text, or even creating slides on the fly:

On the Teams side, you can have simultaneous translation or subtitling or even a summary of the key points of your meeting:

Zoom is trying to keep up with Zoom IQ, an AI-powered tool for brainstorming and meeting summarization.

Additionally, Microsoft Designer, a combination of Dalle and Canva-like features, has been announced for creating images.

A similar visual creation tool should soon be available in Bing search engine :

Microsoft is also rolling out CopilotX for GitHub, a powerful tool for developers that helps them code more quickly and with fewer errors. Welcome to text to code!

In other news, Microsoft has announced Loop, its version of Notion,...

...while Notion has launched Notion AI.

Meanwhile, Google is trying to catch up with its competitors after a disastrous keynote where its stock price dropped by $120 billion due to a false response from its AI during a demo. Since, the two teams from Deepmind and Google Brain had to bury the hatchet and unite against OpenAI.

It's not surprising that Google is one of the players asking for a "pause" on AI, at a time when they are behind and may have even used data from ChatGPT !

BARD, Google's AI similar to ChatGPT, is now launched but still on a waiting list and inaccessible in many countries like France. You can feel that Google/Alphabet is really stressed and behind. Here's a little comparison between the two. On top Google has announced an upcoming AI for its Workspace tools (Gmail, Google Sheet, Slides, Docs...):

Apple has been relatively quiet on the AI front, although there are rumors of a new Siri project codenamed "Bobcat".

We can also expect to see a proliferation of ChatGPT-like applications that are specialized for specific subjects or industries. For example, there's LightOn from Paradigm and Dolly from Databricks, which has 6 billion parameters compared to ChatGPT3's 175 billion, based on Eleuther.

The goal? To offer companies a solution to train on specific topics without having to share their data outside, as with ChatGPT. Some large companies or the Montpellier city hall, Samsung, Orange have already partially or totally banned their employees/agents from using ChatGPT due to privacy concerns.

On the generative AI and graphic design side, innovations are also advancing rapidly.

It will be difficult to list all the practical tools like Ailabtools, Photoroom (amazing for e-commerce visual assets), etc., whose number is exploding every day. So let's focus on the biggest announcements:

Midjourney has deployed its version 5 which offers an incredible level of graphic rendering and solves many previous problems (frequent errors in hands, number of teeth, etc.).

At HUB Institute we are huge fans of the renders from this solution that we have been using with our teams for over 6 months to illustrate trend reports, articles,... As someone who has been doing photography for 10+ years, my business partner, Emmanuel Vivier see less and less difference with the 40 photos that he has created and shared on his Instagram account in the past few weeks using Midjourney:

Canva announced the arrival of many AI features within its platform during its 2nd Canva event:

• Magic design that generates template ideas on the fly
• Canva Presentation to generate a first version of a slide deck from a few keywords.
• Built-in translation to save time
• An improved "Text to Image" feature

• Draw your ideas: to transform a hand-drawn sketch into geometric shapes
• Magic Eraser but especially Magic Edit which allows you to describe in a few words what you want to modify or create in an image like in Dall-e.
• Beat Sync to synchronize your visuals to the rhythm of music

After the success of Dall-e, Midjourney or Stable Diffusion in the past 6 months, Adobe has finally come out with the announcement of Adobe Firefly, its generative AI creation suite. A promising suite whose AI has been trained on the stock image base of the brand (avoiding copyright risks).

Our initial tests show that Adobe Firefly is not yet at the level of Midjourney 5 which has solved many problems with the representation of hands or faces for example:

We will also mention Instruct-NeRF2NeRF which allows you to modify a recorded scene in NERF aka Neural Radiance Fields (via Luma) with simple text!

Text to Video is shaping up to be a near reality

RunwayML, which had already made a lot of noise with Gen1 (now available to everyone!) its alpha tool Text to Video, is back with the announcement of Gen2:

So yes, the rendering is still of low quality, but as Midjourney was 6 months ago! A few more months, new algorithms, more affordable computing power and voila! Gen1 is already available at a rate of 14 credits per second of generated video (1000 credits at $10). We are also witnessing a multiplication of new "Text to Video" tools that are not necessarily very advanced yet, with sometimes quite comical results:

Others like Corridor Digital find ingenious strategies such as video to video. They shoot a scene in low-cost mode on a green screen. They export the sequence image by image to Stable Diffusion, apply a generative AI treatment (here a manga rendering), and voila, after anti-flicker cleaning in Davinci Resolve, they get an animated cartoon 30 times faster and cheaper than with traditional tactics!

Finally, the video game industry has also reacted.

Tools like Stable Diffusion or Scenario.com will allow for a spectacular acceleration of asset creation in video game production.

Nvidia, one of the world leaders in real-time 3D technologies, is working on its Omniverse, a hybrid platform between gaming, industrial digital twin, architectural design, and more.

Meanwhile, Fortnite (with 500 million players) is launching a new creation tool, "Unreal Editor for Fortnite." A brand can easily import its logo or a 3D model of its product (clothing, caps, cars, etc.), effectively turning Fortnite into a much more open metaverse.

Nvidia has also presented Picasso, its cloud dedicated to generative AI applications such as text-to-image, text-to-video, and text-to-3D:

Unity has announced this week the launch of a new program around generative AI for developers. One can imagine that eventually, we will be able to generate 3D models from a simple text query (aka "prompt").

On its side, Ubisoft has presented a Ghostwriter AI to save time writing dialogue for non-playable characters:

And Roblox is launching a tool for creating objects in its metaverse from text. All you have to do is describe an object and its effects, and the code will be generated automatically ??

This news is reminiscent of that of Scenario.com for creating elements for Sandbox using generative AI (Text to Metaverse??)

In short, we are experiencing a real #AIcceleration that should impact a considerable number of professions, both in their processes and sometimes in their reason for being. Some, like Elon Musk and hundreds of world experts, are even calling for a moratorium on research into AIs more powerful than ChatGPT 4...

One thing is certain, as Bill Gates announced in one of his recent posts: "The Age of AI has begun"!

Get your business ready for Generative AI

?? Get your team and execs trained on Generative AI with our experts ??

Let's meet

???Data & AI for Business : April 18-19, 2023 - Pavillon d'Armenonville, Paris - Porte Maillot

???Leaders & Impact Forum : May, 11th, 2023 - 55 avenue Bosquet, Paris, 7e - MEDEF

???Work Experience : June 1st, 2023 - HUB Loft, Paris 8e

The HUB Institute wants you!

?? We are hiring! Are you looking for a job, an internship or apprenticeship in an innovative, digital, sustainable environment??

?? Find out out about our offers on Welcome to the Jungle.

The members of our team will introduce you to the universe of the first French Digital & Sustainable Think Tank, its missions and its opportunities ??

#AI?#IA?#innovation?#generativeAI?#firefly?#copilot?#chagpt?#midjourney?#dalle?#stablediffusion?#AIccélération