📝 Google’s self fact-checking AI

ALSO: From $150M to $1B valuation, Grok’s image generator

Estimated Read Time: 2 to 3 minutes

Ever used AI to complete a project only to find it makes up crap or is downright wrong? It can seriously reduce productivity when AI needs to be fact checked for inaccuracies. It defeats the point of using AI. And Google has a novel solution called SCoRe to fix this. Let’s dive in!

• 🧠 Google introduces AI self-correction.

• 📸 Grok’s image generator gets $100M funding at $1B valuation.

• ⚡️ Microsoft taps into Nuclear power.

🧠 Google introduces AI self-correction

What happened: Google DeepMind revealed a new approach called Self-Correction via Reinforcement Learning (SCoRe) allowing LLMs to autonomously correct their errors without outside supervision.

The details: 

• Google DeepMind’s SCoRe enables LLMs to use self-generated data through multi-turn reinforcement learning so they can fact check themselves.

• This is a big deal because most LLMs make stupid mistakes all the time. Meaning you need some method to fact check them.

• Previous approaches often rely on manual fine-tuning or external verifier models, which introduce a whole load of problems themselves. (Basically they were damn expensive and unreliable.)

• The model goes through two stages of training: the first makes big changes, the second focuses on rewarding improvements. (Rewarding good actions is the basis of learning models.)

• This helps ensure the model focuses on making accuracy improvements rather than changes for the sake of change.

• Real world testing: Applied to Gemini 1.0 Pro and 1.5 Flash models, SCoRe achieved a 15.6% improvement in mathematical reasoning tasks and a 9.1% improvement in coding tasks.

• More results: The models’ accuracy increased from 60.0% on the first attempt to 64.4% on the second. So yes, it is effectively learning.

Why it matters: For a while now, AI experts and CEOs have been saying that self-fact-checking AI are the next big step (along with full blown agents). This is one of the biggest problems with current AI models; you need to fact check them. AI is about to become a lot more reliable.

New Weekly Featured Articles Coming!

To add even more value to our newsletter, we’ll be adding new article types and features to the newsletter!

AI is constantly evolving and changing. There’s a LOT of tools out there. It’s a mess, honestly. So next newsletter we’ll start covering the best AI tools across various categories, starting with Chatbots, Image Generators and Video Generators. After that we’ll move to other productivity tools to help simplify this whole mess of tools.

We’ll also start covering how corporations plan to use AI in the workplace. Plus other topics like the basics of LLMs, how the job market is tuned to shakeup, deep dives into specific models and tools, covering various workflows and more. Stay tuned!

🎶 Indian filmmaker Ram Gopal Varma says he will exclusively use AI-generated music in his future projects, including his new movie “Saree,” whose entire background score is AI-produced. He launched RGV Den Music to feature music from AI apps like Suno and Udio. The reason? It’s easier and cheaper of course! So Hollywood is not alone, even Bollywood is changing to use AI. It’ll be the same story when video generation advances enough.

🤔 The FTC thinks tech companies have too much info on you. A new FTC report scrutinizes how major social media and streaming companies collect and monetize user data, revealing that firms like Amazon, Facebook, and ByteDance provided evasive responses about their practices. This comes as no surprise, but what matters are the consequences. The FTC is planning new regulations concerning privacy (especially for children). We could see a serious shakeup in laws (they’ll probably lobby the hell out of this though).

🎥 Amazon released a video generator only for ads. Amazon has unveiled Video Generator, an AI-powered tool that transforms a single product image into a promotional video, currently in beta for select U.S. advertisers. It uses a single product image to create a video ad for the product. The biggest criticisms are about copyright infringement. I have no doubt it’ll be possible to falsely represent products with this tool.

📸 Grok’s image generator gets $100M funding at $1B valuation

What happened: Black Forest Labs, a generative AI startup specializing in image models, is reportedly closing a $100 million funding round at a $1 billion valuation. Co-founded by engineers behind Stability AI, the company has attracted significant investor interest and secured Elon Musk’s xAI as a major customer.

The details:

• Black Forest Labs is in the process of closing a $100 million funding round at a $1 billion valuation, though the deal is not yet finalized. Last time they were valuated was at $150M. That’s a big increase.

• The startup was co-founded by Andreas Blattmann, Patrick Esser, Dominik Lorenz, and CEO Robin Rombach; all researchers who previously developed technology for Stability AI.

• Elon Musk’s xAI is using Black Forest’s Flux 1 model to power image generation in its Grok chatbot. So they have proven results.

• Their next venture is a text-to-video model. A common move for image generators.

Takeaway: The amount of money being thrown around is crazy. Deals like these show faith in the product and its future. There’s a ton of interest growing in image generation, especially the uncensored kind. I fully expect Black Forest Labs to be an influential mover in the future. There’s too many important partnerships for it to fail at this point.

⚡️ Microsoft taps into Nuclear power

What happened: Microsoft has signed a significant deal with Constellation Energy to purchase power from the Three Mile Island Unit 1 nuclear plant to meet the immense energy demands of its AI data centers.

The details: 

• Microsoft will buy all the power generated by the revived Three Mile Island Unit 1 reactor over the next 20 years. The reactor will provide 835 megawatts once active.

• And they’re not alone. Amazon and OpenAI are also turning to nuclear power. Sam Altman himself has stated nuclear power is necessary to develop better AI.

• The Scale: Upcoming data centers could consume 508 terawatt hours annually, exceeding Australia’s total electricity production. Let that sink in.

• Side note: The project is expected to create 3,400 jobs, add $16 billion to Pennsylvania’s GDP, and generate over $3 billion in taxes. f

Why this matters: The biggest obstacle for AI companies are the energy costs. In fact, much of the world has trouble meeting common energy demands. AI data centers only add to that burden. Expect energy companies to skyrocket in value and become a major point of contention in the future.

🍰😋 Dessert is served!

Source: Midjourney

Give us your feedback!

Got feedback? Rate today’s newsletter by clicking below! 

Got more personal feedback? Fill in the box when you rate us and give us your criticism and feedback.

Thank you for reading! 

❤️Share the (ARTIFICIAL) Love!

Got a friend who wants to appear as smart as you? An AI fanatic or someone shaking in their boots of getting replaced? Share us by clicking the link below and you’ll get our love ;)