- AI Fire
- Posts
- š¤ Gemini 2.5 Smashes Arena with +40
š¤ Gemini 2.5 Smashes Arena with +40
Gemini can see what you see?

Plus: Gemini can see what you see?
Read time: 5 minutes
Just one day - and already flooded with new AI models and features. Itās a tech showdown between OpenAI and Google and a global race between the U.S. and China.
What are on FIRE š„
IN PARTNERSHIP WITH PERSPECTIVE AI
Use AI to have real customer conversations, uncover deep insights, and make smarter business decisionsāstart free today!
Understand what truly drives customer decisionsābeyond surveys and guesswork. Perspective AI scales real, in-depth customer conversations, helping you uncover hidden motivations, predict needs, and act with confidence.
š Try it freeāyour first 10 conversations are on us!

AI MODEL
š§ Gemini 2.5 Pro Is Here - And It Actually Thinks Before It Speaks

Google DeepMind just dropped its smartest AI yet - Gemini 2.5 Pro - and it's not just about speed or scale anymore. This thing reasons. Like, really thinks through stuff before replying.
Itās a āthinking modelā, trained with techniques like reinforcement learning and chain-of-thought prompting, meaning it doesnāt just spit out predictions ā it works through the logic first.
Itās already beating GPT-4.5 and Claude 3.7 Sonnet in math, science, and coding tasks.
Scored 63.8% on SWE-Bench Verified, which basically means itās crushing complex code tasks from minimal input. (Yes, it can build a video game from a single line.)
Comes with a 1 million token context window (2M on the way!) and handles everything ā text, video, images, code ā in one go. No duct-taped multimodal hacks here.
No expensive ātest-time tricksā needed. It wins on clean performance.
Now #1 on the Arena leaderboard.
Why it matters: Gemini 2.5 is different: itās designed to reason, step by step, more like a human would. It doesnāt need workarounds like āmajority votingā to perform well - which also means itās more efficient and scalable for real use.
This model could be a foundation for the next generation of AI agents - ones that actually understand what theyāre doing, not just mimicking patterns. And with long context, native multimodal input, and top-tier coding skills, itās built for serious work, not just fancy demos.
PRESENTED BY SALEFORGE
The first search engine for leads
Leadsforge is the very first search engine for leads. With a chat-like, easy interface, getting new leads is as easy as texting a friend! Just describe your ideal customer in the chat - industry, role, location, or other specific criteria - and our AI-powered search engine will instantly find and verify the best leads for you. No more guesswork, just results. Your lead lists will be ready in minutes!
TODAY IN AI
AI HIGHLIGHTS
š¼ļø OpenAI just launched a new image generation model, fully built into GPT-4o and Sora. It's their best yetāprecise, accurate, and stunningly photorealistic.

š DeepSeek just launched DeepSeek-V3-0324 on Hugging Face! With 37B active parameters per token, it runs smoothly even on a Mac Studio - no data center needed.
š§ Qwen just launched Qwen2.5-VL-32B-Instruct. Despite being smaller, it beats models like Mistral-Small-3.1 and even the 72B version of Qwen2. Smart fine-tuning and RLHF, it delivers high performance without heavy compute - great news for devs who want power without the cost.
š¼ Deloitte just launched AI Advantage for CFOs - a smart new platform built with AWS and Anthropic. It helps finance teams automate work using AI agents that can understand docs, images, audio, ā¦ tailored for CFOs.
š® Nvidia has released G-Assist, a gaming-focused AI chatbot that runs entirely on your GPU, without needing the cloud. Itās an experimental tool designed to optimize gaming performance even adjust hardware like overclocking or RGB lighting.
š ļø Pieter Levels, 38, just built a $1M business in 17 days using AI tools and zero code. His latest project, fly.pieter.com, is a browser-based flight simulator made in just 3 hours with tools like Three.js and Cursor. Itās now pulling in $87K/month!
š° AI Daily Fundraising: Nexthop AI just raised $110M to supercharge cloud networking for AI. Backed by Lightspeed and others, theyāre building custom hardware and software to power hyperscale AI clusters faster and smarter.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
š¦¾ Alibabaās LHM is a new AI model that creates animated 3D avatars from just a single reference image.
š¢ BASE44 2.0 is the all-in-one app building platform. No integrations needed.
šØš»āš» LiftmyCV is AI agent that finds jobs and auto-applies on your behalf
šÆ Lamatic 2.0 is IDE to build & deploy AI agents on serverless
š¬ Krea Video Training trains AI to create videos in your style
AI QUICK HITS
š¤ NVIDIA introduced Isaac GR00T N1 ā First Open Humanoid AI Model (Link)
šµ Anthropic Scores Early Win in AI Music Copyright Battle (Link)
šØāš©āš§ Character.ai Lets Parents See Kids' Bot Chats (Link)
šø Gemini Can āSeeā Screens and Camera Feeds in Real-time (Link)
š OpenAI Reshuffles Leadership as Sam Altman Shifts Focus to Research (Link)
AI INSIGHTS

š§ Howās Your āAI Brainā Level? |
We offload a lot these days - calendars, reminders, GPS, writing help, even decision-making. It's efficient. But itās also quietly changing how our brains work.
Weāve always used tools to think better: writing, calculators, sticky notes. But now? Itās not just help - itās replacement. And thatās where things get tricky.
What does the science say?
A growing pile of studies is raising red flags. When we rely on AI too much, we lose critical thinking muscle. Memory, focus, and problem-solving start to weaken. One study found that when people used reminders, brain activity in planning areas dropped significantly.
Our brains process ~10 bits per second ā slower than a 1960s modem.
Tools like ChatGPT, Google, and smart assistants help us move faster.
But when we use them too much, we lose something: deep thinking.
āOnce you know information is stored outside your brain, you stop holding onto itā
That sounds efficient. But it's also how you forget your own phone number.
So what's actually changing?
Kids now struggle with attention because of constant short-form content.
Even experienced workers say AI is making them faster, but not sharper.
Some people now trust AI answers without verifying. Thatās not helpāthatās dependency.
Are we becoming cognitive misers?
Maybe. It's tempting to save brainpower, especially when everything around us moves fast. But itās worth asking: What are we giving up when we stop wrestling with hard problems?
A quote that stuck with me:
āSolving something on your own can give you insights youād never get from an answer handed to you.ā
Thatās the thing. AI might get you to the answer, but it robs you of the aha moment.
So whatās the move here?
We donāt need to ditch AI. But we do need to think about what kind of thinking is too important to give up. Not every problem needs to be automated. Some are worth the effort. What do we keep doing ourselves?
Read long articles (yes, like this).
Solve hard problems without AI first.
Write, think, reflect - before you outsource it.
Because once you give away the thinking... itās hard to get it back.
Final thought:
AI should be your tool, not your brain. Keep the good parts of thinking - curiosity, struggle, discovery - alive. Use the tech, but donāt become it.
AI JOBS
We read your emails, comments, and poll replies daily
How would you rate todayās newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello ā we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply