🚨 Stability AI steals Midjourney data

SIGN UP HERE

Welcome, AI RESEARCHERS.

AI Betrayal: Midjourney Severs Ties with Stability AI After Shocking Incident

Plus, OpenAI fires back at Elon & huge developments from Midjourney, Amazon, Google, ebay and more,

Here the 10 most important things happened:

In today’s AI Researchs:

• Stability AI grabs Midjourney data

• OpenAI fires back at Musk with secret emails

• Whistleblower warns about Copilot Designer’s risks

• Try out Google's most recent LLM using your phone

• 6 new top AI tools

• More AI Trending Agendas & News

• New Ai Research papers (Education)

Read time: 5 minutes

AI DRAMA

🚨 Caught Red-Handed: Midjourney Exposes Stability AI's Data Grab

AI Researchs: According to a Midjourney office hours session, an individual at Stability AI allegedly attempted to collect large amounts of prompt and image data on a Saturday night, causing a service disruption

The important details:

• Saturday Night Outage: A significant service outage occurred on Midjourney's platform on Saturday night, allegedly triggered by the actions of a Stability AI employee.

• Malicious Intent Alleged: Midjourney asserts that the activity was a targeted and deliberate attempt to access their data, not an accidental occurrence.

• Swift Ban Enacted: All Stability AI employees have been banned indefinitely from using Midjourney's services in response to the incident.

New upcoming features 

Drama aside, Midjourney's dropping some killer new features.

• Character Consistency (this week): Ensures characters look the same across generations.

• Enhanced Describe (next week): Improves accuracy in translating text descriptions to images.

• --style random for --v 6 (2 weeks): Adds style variety within Midjourney's Version 6.

• Style Explore (2 weeks): Allows deeper investigation of artistic styles.

OpenAI

📢 OpenAI fires back at Musk with secret emails

AI Researchs: OpenAI hits back at ex-co-founder Elon Musk’s legal action, publishing a blog.post with a series of his own emails to counter his allegations that the company strayed from its original vision for the sake of profits.

The important details:

• Emails reveal Musk pushing OpenAI to raise at least $1B, urging the nonprofit to find more funding to pursue its ambitious AGI objectives.

• Musk also proposed in 2018 that Tesla should buy OpenAI as "the only path" for the startup to remain competitive and beat Google.

• Another notable point is Ilya Sutskever, who appeared under the blog’s authors after being out of public sight since November’s drama.

• Sutskever stated in one email that the ‘Open’ in OpenAI means everyone gaining from AI, not sharing the science — sparking controversy in the open-source AI community.

• Musk reacted to the blog by tweeting that he would withdraw the lawsuit if the company renames itself to ‘ClosedAI’.

Why it matters: The saga around OpenAI keeps being the most entertaining show in tech, and Elon’s latest appearance adds a whole new level of zest. While the emails show a different perspective, Musk is not one to give up on a fight — implying there are more likely more explosions to come.

MICROSOFT

🚨 Whistleblower warns about Copilot Designer’s risks

AI Researchs: A Microsoft senior software engineer reported Copilot Designer to the FTC, claiming the company is failing to protect the AI from creating harmful content.

The important details:

• Shane Jones found a flaw in OpenAI’s DALL-E (which Copilot Designer uses), that let him evade content filters.

• Jones said he asked the company to withdraw the product until more security measures were implemented, but was dismissed.

• In his letter, Jones accuses the tool of showing graphic violence, underage drug/alcohol use, political bias, copyright violations, and more.

Why it matters: This shocking allegation from a Microsoft insider seriously challenges the tech giant’s public pledge to “responsible AI” principles — while also implying profits may be overriding safety.

Ai TRAINING

📱 Try out Google's most recent LLM using your phone

• AI Researchs: This tutorial will teach you how to run Google’s latest Gemma 2B model on your phone, which allows you to have privacy and no internet connection.

Step-by-step:

For iPhone users, there is a beta app you can use.

1. Go to this link, where you will find two steps. First, pick step 1, “View on App Store,” and download Testflight.

2. Go back to the website and click on step 2, “Start Testing” to install MLC-Chat.

3. Once you install the app, press the “Cloud” icon next to the Gemma 2B model to download. You can also download and experiment with other models such as Mistral 7B Instruct or Phi-2.

Wait for a few seconds, and you will be able to chat with it.

For Android users, you can also find a demo app, where you will have to download the APK file. videos if you need more help.

RESEARCH OF THE DAY

🪖 The U.S. Army is testing GPT-4 in advisory roles for warfare

AI Researchs: The US Army’s Research Lab is supposedly testing with commercial AI systems such as GPT-4 for strategic planning roles in military simulation exercises.

The important details:

• During a trial, AI-driven chatbots developed strategic plans for a combat scenario within the StarCraft II game environment.

• Among the AIs tested, OpenAI’s GPT-4 demonstrated superior performance, despite not being flawless.

• The Department of Defense has pinpointed an extensive list of 180 potential applications for AI in military contexts, collaborating with firms like Palantir and Scale AI for technological integration.

• Specialists in the field advise caution in the military application of AI, pointing out issues related to practicality and the potential for ethical and bias-related dilemmas.

Why it matters: AI's role in warfare, while essential, sparks ethical debates. Amidst a competitive AI arms race, nations may rush its military use, readiness aside.

NEW AI TOOLS:

• 📊 Profit Leap: A tool for business intelligence and strategic insights.[LINK]

• 💡 IdeaSpark: A tool to generate multilingual personalized business ideas. [LINK]

•  📓 Vocalo: A tool to improve English speaking skills through conversational practice. [LINK]

• 💭 RambleFix: A tool to convert spoken thoughts into written text.[LINK]

• 🧾 v0 report: A tool to automatically generate business reports. [LINK]

• 📽️ SDXL Lightning: A fast text-to-image generation model that produces high-quality images. [LINK]

AI AGENDAS

• Palantir secured a $178M U.S. Army contract to build 10 TITAN vehicle prototypes. This AI-powered technology will enhance battlefield intelligence, providing soldiers with faster, more accurate decision-making data.

• Alibaba led a $600 million funding round for Chinese AI startup MiniMax, their second major AI investment this year.

• Amazon continues to invest in a greener future with a $53M commitment to Glacier, an AI startup tackling the challenges of waste management and recycling efficiency.

• Google upgrades its search algorithm to fight AI spam and improve results, with the goal of slashing low-quality content by up to 40%.

• eBay enters the LLMs arena, developing its own LLM trained on e-commerce data and ramping up its GPU capacity for accelerated growth.

• U.K. doctors now have a powerful AI tool that predicts kidney failure at unprecedented speeds, leading to faster interventions and improved patient outcomes..

• Microsoft's defense of OpenAI likens the NYT's copyright lawsuit to the Hollywood industry's failed opposition to VCRs. They argue the NYT is trying to hold back technological progress.

Ai Education

Design2Code 

How Far Are We From Automating Front-End Engineering?

New research shows AI is getting surprisingly good at turning design mockups into working code. This Design2Code breakthrough highlights how AI could save developers loads of time. Fine-tuned open-source models are even showing promise alongside premium AI tools. 

Caduceus

Bi-Directional Equivariant Long-Range DNA Sequence Modeling

Caduceus is a new type of DNA language model excelling at understanding complex, long-range interactions within DNA sequences. It accounts for the unique properties of DNA, including directionality and reverse complementarity.

MathScale

Boosting Large Language Models for Math with MathScale

This research introduces MathScale, a method to create a massive dataset of math questions and answers to enhance the capabilities of large language models (LLMs) in solving mathematical problems. Inspired by how humans learn math, MathScale extracts topics and knowledge points from existing problems and builds a concept graph to generate new questions..