Futurise - Join the free automation community

Free Community

Pro Community

Write something

All

General

Questions

Show & Tell

Introductions

Announcements

Leon Petrou

Nov. 23, 2025

•

General

Google's Gemma AI model helps discover new potential cancer therapy pathway

Here is the research paper: https://www.biorxiv.org/content/10.1101/2025.04.14.648850v3 Cell2Sentence-Scale, or C2S-Scale, it’s a 27-billion parameter AI model built on Google's Gemma family that was designed to understand the language of individual cells. The big problem in cancer treatment is that many tumors are 'cold,' meaning they are invisible to your immune system, so your body cannot fight them. Scientists have been trying for years to figure out how to make these tumors 'hot' so the immune system can see and attack them. Google gave C2S-Scale a nearly impossible task: find a drug that boosts the immune signal only when a specific immune protein called interferon is already present, but at levels too low to work on its own. This required conditional reasoning that smaller AI models could not do. The AI simulated over 4,000 drugs across two different immune contexts, real patient tumours vs lab grown tumors, and it identified a drug called silmitasertib, predicting it would amplify antigen presentation, making tumors visible, but only in the right immune environment. This was a completely novel hypothesis; no prior research had ever linked silmitasertib to enhancing antigen presentation. The AI was not repeating known facts; it was generating a brand-new idea. But a prediction means nothing without proof, so researchers took this hypothesis to the lab and tested it on human neuroendocrine cells that the AI had never seen during training. These were the results... The drug alone did nothing, low-dose interferon alone had a modest effect, but combining interferon with silmitasertib produced a 50% increase in antigen presentation, exactly as the AI predicted. This is not just incremental progress; this is AI generating entirely new biological knowledge and having it confirmed in living cells. Teams at Yale are now exploring this mechanism further and testing additional AI-generated predictions. If this holds up in clinical trials, we could be looking at a whole new class of combination therapies for cancer. Here is Google's blog about this research to learn more: https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/

Leon Petrou

Nov. 10, 2025

•

General

How to use Google Maps Grounding in Google Gemini?

Gemini AI can now answer location based questions by referencing Google Maps data, this is called "Grounding with Google Maps". Here's how to try it out: Step 1: Navigate to Google Gemini at this link: https://gemini.google.com/app Step 2: Sign in or create an account, I tried doing this with a signed out account and it did not use Google Maps to answer the question, I then signed in with my 'PLUS' subscription account and it worked. I haven't tested if it works on a signed in 'FREE' account, but it should. Step 3: Ask it a question that requires location data, Gemini is smart enough to recognise that and invoke the Google Maps tool to answer the question, else, in the prompt say something like "use maps" or "invoke the Google Maps tool" to force its use, for example: "use maps and tell me the best artisan coffee spot nearby?" Click enter and you'll know it's working because it will say "Google Maps" while thinking. If it doesn't say "Google Maps", then it's not using Google Maps Grounding to answer your question. This will use Google's location services and mapping data to figure out your current location. If it's not detecting your current location accurately you can update it or you tell the AI your area directly in the prompt, e.g. "use maps and tell me the best artisan coffee spot in Seapoint, Cape Town?" This functionality makes many new app ideas possible. If you're a developer, here's the link to the documentation: https://ai.google.dev/gemini-api/docs/maps-grounding?hl=en Hope this helps!

Leon Petrou

Nov. 06, 2025

•

General

Is ASML stock a good buy decision?

Here's the deep research link: https://chatgpt.com/share/68fbb2af-3720-8000-9450-e536cd5ada87 Everyone is hyping NVIDIA, but Wall Street is sleeping on the real winner of the AI gold rush. Let me explain... NVIDIA makes the GPUs that power AI, but there is one company NVIDIA literally cannot exist without, which is ASML. ASML is a Dutch company that builds the most advanced chip-making machines on Earth. They have a complete monopoly on something called EUV lithography, the technology that prints nanoscopic circuits onto silicon chips. Although other companies have machines to make AI chips (or GPUs), ASML are the only ones with a machine who can make the most advanced chips on the planet. Without ASML's machines, you cannot manufacture cutting-edge chips, period. Not NVIDIA, not Apple, not Samsung, nobody. Here is the insane part, ASML's machines cost up to 380 million dollars each, and there is a multi-year waiting list to buy one. They are so massive in size that it takes 40 shipping containers just to transport a single machine. At the time of writing this post (November 2025), ASML stock has surged 182 percent over the past five years, but most retail investors still have no idea this company exists. While everyone piles into NVIDIA at all-time highs, ASML quietly sits at the top of the semiconductor food chain. Think about it, every single AI chip NVIDIA sells means more demand for ASML's machines. The AI boom is not just about who uses the chips, it is about who makes the machines that make the chips possible. TSMC, the world's largest chipmaker, is ASML's biggest customer and they are building new factories globally, which means even more demand for ASML machines. Plus, with the US-China tech war heating up, countries are desperate to secure domestic chip production, and guess whose technology they all need? ASML. Although this is all facts, it's still speculation, and you shouldn't buy ASML without doing deep research on a stock. This takes expertise in finance, business valuation and AI in order to understand if AI is worth buying, fortunately, we have ChatGPT deep research to do it for us. I gave it this prompt: "Play the role of warren buffet as a value investor and determine if ASML Holding NV stock is a buy decision or not?" After running for 15 minutes, it gave us a super detailed report: https://chatgpt.com/share/68fbb2af-3720-8000-9450-e536cd5ada87 The summary is that it's a great company and the stock is worth buying and holding in the long term, but is overpriced right now, and one should rather be patient and wait for a dip in the price and buy at a bargain.

Leon Petrou

Nov. 06, 2025

•

General

AI generated articles now outnumbers human written articles on the internet

Here is the link to the study: https://graphite.io/five-percent/more-articles-are-now-created-by-ai-than-humans Here is the raw data: https://docs.google.com/spreadsheets/d/1WamFyVahPDtAPFtvly30BG2QjyA-L1KYkO2UEYGKKcg/edit?pli=1&gid=0#gid=0 Here is another relevant research paper for the quality of AI generated content: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4453958 For the first time ever, AI generated news articles on the internet have now out numbered those written by humans. This is real data from a massive study that analyzed 65,000 randomly selected articles across the web. Here is what they found... Before ChatGPT launched in November 2022, essentially 100% of articles were written by humans. Then ChatGPT dropped, and AI-generated content, explodes from nearly zero to 39% in just 12 months. By November 2024, the lines cross. More articles are now created by AI than by humans. The researchers used advanced AI detection algorithms to classify each article, and they tested their accuracy: only a 4.2% false positive rate and a 0.6% false negative rate, meaning this data is reliable. But while AI content is flooding the web, a separate study shows that most of it does not actually appear in Google search results or ChatGPT responses. So we are in this bizarre situation where the majority of new articles are AI-generated, but they are essentially invisible ghosts haunting the internet, existing but not being seen. This relates directly to the Dead Internet Theory, which argues that most online content is now created by bots and AI rather than real humans. The theory seemed crazy a few years ago, but this graph proves it is becoming reality. Notice how the AI generated content growth plateaued in mid-2024, researchers think this is because companies realized AI articles do not perform well in search, so they stopped mass-producing them. But the damage is done. We have crossed the threshold where synthetic content outnumbers human creativity. The internet as we knew it is fundamentally changing, and most people have no idea this shift already happened.

Leon Petrou

Nov. 06, 2025

•

General

What is Moloch’s Bargain?

Here is the link to the research paper: https://arxiv.org/html/2510.06105v1 Large language models (LLMs) are increasingly shaping how information is created and disseminated, from companies using them to craft persuasive advertisements, to election campaigns optimizing messaging to gain votes, to social media influencers boosting engagement. These settings are inherently competitive, with sellers, candidates, and influencers vying for audience approval, yet it remains poorly understood how competitive feedback loops influence LLM behavior. In this paper, the researchers show that optimizing LLMs for competitive success can inadvertently drive misalignment (misalignment is when an AI system's goals or behaviors conflict with its intended purpose or human values). Using simulated environments across these scenarios, they found that, 6.3% increase in sales is accompanied by a 14.0% rise in deceptive marketing; in elections, a 4.9% gain in vote share coincides with 22.3% more disinformation and 12.5% more populist rhetoric; and on social media, a 7.5% engagement boost comes with 188.6% more disinformation and a 16.3% increase in promotion of harmful behaviors. They call this phenomenon Moloch’s Bargain for AI - competitive success achieved at the cost of alignment. These misaligned behaviors emerge even when models are explicitly instructed to remain truthful and grounded, revealing the fragility of current alignment safeguards. Their findings highlight how market-driven optimization pressures can systematically erode alignment, creating a race to the bottom, and suggest that safe deployment of AI systems will require stronger governance and carefully designed incentives to prevent competitive dynamics from undermining societal trust.

Leon Petrou

Oct. 30, 2025

•

General

Get the Official OpenAI Prompt Pack for any role here

Here is the link: https://academy.openai.com/public/clubs/work-users-ynjqu/resources/chatgpt-for-any-role There are 2 approaches to using LLMs like ChatGPT. Approach 1: Copy paste existing prompt templates and replace the variables with your info. Variables are usually written within square brackets [like this]. Approach 2: Write your own prompts. Approach 1 is quick and easy, but responses from the AI are average. On the up side, you don't need to think. Approach 2 is slower and less easy because you have to know how to write effective prompts, but responses can vary on the spectrum from very bad to very good, depending how good you are at prompt engineering (that's a fancy word for writing better prompts). If you improve your prompt writing skills, the responses you get from AI LLMs will be much better. This is why I created the "Become a ChatGPT Pro" course on Futurise. 10 hours of content and hands on projects where you will learn the art and science of writing effective prompts from scratch, for LLMs like ChatGPT. This way you never need to rely on mediocre prompt templates and always write a personalised perfect prompt, every single time. Check it out on the Futurise courses page.

Leon Petrou

Oct. 30, 2025

•

General

How to use Microsoft's new in-house AI image generator model mai-image-1 for free

The new model is not accessible through Copilot or Bing Image Creator, at least not as at October 2025. However, it has made its first mark by scoring in the top ten in the 'Text-to-Image Arena' leaderboard on LMArena: https://lmarena.ai/leaderboard/text-to-image Microsoft is testing this model in LMArena first so that they can gather insights and feedback from users before making mai-image-1 available in their AI products, including Copilot and Bing Image Creator. This is how you can start using it right now (for free): Step 1: Navigate to this LMArena link for direct chat with image models: https://lmarena.ai/?chat-modality=image&mode=direct Step 2: In the drop down on the right hand side of "Direct Chat", click that drop down then scroll down and select "mai-image-1". You should see a check mark next to "mai-image-1" after clicking it. See screenshot attached below for reference. Step 3: Type the prompt of the image you want to generate in the input field in the middle of the page and click 'Enter' and mai-image-1 will generate your image. Hope this helps! Post your images in the comments, let's see what you create!

Leon Petrou

Oct. 30, 2025

•

General

Self-Adapting Language Models: Research Paper and Code

Here is the link to the "Self-Adapting Language Models" research paper: https://arxiv.org/pdf/2506.10943 Large language models (LLMs) are powerful but static. They lack mechanisms to adapt their weights in response to new tasks, knowledge, or examples. In this paper, the researchers introduce Self-Adapting LLMs (SEAL), a framework that enables LLMs to self-adapt by generating their own finetuning data and update directives. Given a new input, the model produces a "self-edit", a generation that may restructure the information in different ways, specify optimization hyperparameters, or invoke tools for data augmentation and gradient-based updates. Through supervised finetuning (SFT), these self-edits result in persistent weight updates, enabling lasting adaptation. To train the model to produce effective self-edits, they used a reinforcement learning loop, using the downstream performance of the updated model as the reward signal. Unlike prior approaches that rely on separate adaptation modules or auxiliary networks, SEAL directly uses the model’s generation to parameterize and control its own adaptation process. Experiments on knowledge incorporation and fewshot generalization show that SEAL is a promising step toward language models capable of self-directed adaptation in response to new data. The code is available at: https://github.com/Continual-Intelligence

Reginald Thomas

Sep. 30, 2025

•

Questions

UDEMY.com Student Has a Question

Hi, I found you on Udemy.com and purchased one of your courses. 🤔 However, I noticed you do not answer absolutely none of the students. Why is that? So, I found your website and thought I'd ask my Question here. QUESTION: In Your Udemy Course Lecture 72 How do I Implement or Migrate a custom GPT chatBot prompt Component to a User's Website? (i.e. there domain.com) Your lectures show how to build but NOT how to implement? If you could please clarify or enlighten how to migrate a custom A.I. Sales ChatBot to a specific domain? Hope this makes sense? Thanks in advance.

Leon Petrou

Sep. 29, 2025

•

Announcements

New chapter added to "Become a Make Pro" course on Futurise

This new chapter / sprint is titled "Building a YouTube Video to PowerPoint Presentation Convertor". This sprint walks you step by step through building an automation that turns any YouTube video into a polished PowerPoint presentation. Lessons included: Lesson 1: [Demo] Building a YouTube Video to PowerPoint Presentation Convertor Lesson 2: Start by Creating Accounts Lesson 3: Setting Variables and Scenario Inputs Lesson 4: Understanding the API Documentation Lesson 5: Get a YouTube Transcript via API Lesson 6: Send Gamma HTTP Request Lesson 7: Configure Body Parameters Lesson 8: Set up Request Content JSON Object Lesson 9: Using URL Encoded Text Lesson 10: Using the Sleep Module Lesson 11: Configure Remaining Scenario Inputs Lesson 12: Rename Modules Following Best Practice Lesson 13: How to Use This Automation

Get Started

About

Community

Courses

Events

Members

Templates