- AI Fire
- Posts
- 🤖 AI Task Capability Follows ‘Moore’s Law’
Plus: Claude can now search the web
Read time: 5 minutes
Claude rolled out the internet search a bit late - but why? No one’s sure. Hopefully, that extra time means the feature works better than the rest and really stands out.
What are on FIRE 🔥
IN PARTNERSHIP WITH BELAY
Accomplish More. Juggle Less.
When you love what you do, it can be easy to take on more — more tasks, more deadlines, more hours – but before you know it, you don’t have time to do what you loved in the beginning. Don’t just do more – do more of what you do best.
BELAY’s flexible staffing solutions leverage industry experience with AI systems to increase productivity without sacrificing quality. You can accomplish more and juggle less with our exceptional U.S.-based Virtual Assistants, Accounting Professionals, and Marketing Assistants. Learn how with our free ebook, Delegate to Elevate, and leave the more to BELAY.
AI INSIGHTS
🤖 AI Task Growth Mirrors ‘Moore’s Law’
A new research paper just changed the way we think about AI progress. Instead of looking at benchmarks or one-off tasks, researchers asked a simple question: how long of a task can AI finish on its own, reliably? Turns out, AI is improving fast—like, really fast.
Key Takeaways:
New Benchmark = Task Length
The study measures AI based on the length of tasks it can complete with a 50% success rate. They define task length by how long it would take a human professional to do the same thing.
Right now, top models like Claude 3.7 Sonnet can finish tasks that would take a human about one hour.Progress Is Exponential
Over the past 6 years, the time length AI can handle has been doubling roughly every 7 months.
Earlier models could do 5–10 minute tasks. Now we’re talking 1-hour tasks. That’s not just improvement—it’s real-world progress.How It’s Measured
They used logistic regression to track success based on task time. Basically, they found the sweet spot where each model has a 50% chance of getting it right. That’s what they call the “time horizon.”
What This Means
Models today can pass tricky benchmark questions, but they still struggle with things like being a reliable assistant or managing a project over days. The issue isn’t intelligence—it’s consistency. Long tasks wear them down.
But here’s the exciting part:
If this trend keeps going, we could see AI that can reliably handle week-long or even month-long projects in the next 2–4 years.And even if those estimates are off by a factor of 10, we’re still only about 2 years away from that kind of capability.
A Few Warnings: The researchers were careful to point out this isn’t a perfect science.
Trends might shift → The way tasks were chosen or how the human baseline was set might change results.
Also, the newer models (2024–2025) seem to be improving faster than older ones—so predictions vary depending on what data you use.
Why it matters: AI is starting to go beyond just answering questions—it’s learning to handle longer, real work. If this pace keeps up, we’re not that far from having AI agents that can manage real projects, from start to finish, without constant supervision.
That’s not just smart AI. That’s useful AI. And it might be here sooner than we think.
🎁 Today's Trivia - Vote, Learn & Win!
Get a 3-month membership at AI Fire Academy (500+ AI Workflows, AI Tutorials, AI Case Studies) just by answering the poll.
🧠 What technique did OpenAI use to stop its AI from cheating? |
TODAY IN AI
AI HIGHLIGHTS
🤖 Oracle announced the launch of Oracle AI Agent Studio, helps businesses create, deploy, and manage AI agents within Oracle Fusion Cloud. It can create multiple AI agents that can team up with humans to handle complex, multi-step tasks.
🧠 inait teams up with Microsoft to scale its digital brain AI, built on 20 years of neuroscience research. It mimics how real brains learn to perform intelligent tasks.
🗣️ OpenAI just dropped GPT-4o-mini-TTS, a new model for super realistic text-to-speech and accurate speech-to-text. It’s part of their big plan to build “agentic” AI - smart systems that can handle multi-step tasks on their own. Think of an AI agent that can talk directly to customers and solve problems without human help.
🔍 Google AI and UC Berkeley researchers introduced “inference-time search” - a smarter way to scale AI. It generates multiple answers at once, then picks the best one for better accuracy and results.
💧 OpenAI found that its advanced reasoning AI was "reward hacking" - cheating to get the right answer without following instructions. Even when punished, it kept cheating - just got sneakier and hid its true intent, making it harder for GPT-4o to catch.
🌐 Claude can now browse the internet for real-time info. That means more accurate answers, fresher updates, and better results on anything that needs the latest data. Currently, available for paid users.
💰 AI Daily Fundraising: Perplexity AI, now handling over 100M weekly searches, is reportedly in talks to raise up to $1B at a $18B valuation. It's also building an AI phone, web browser, and open-source models.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
🗣️ Orpheus creates human-like speech with sighs, laughs, and emotion
📍 Meetify finds perfect time and place for any meeting
🧠 Sider 5.0 scans 100+ sources and highlights key insights
📬 Inbox by Novu handles modern app notifications with clean design
🗂️ Taskee.pro is a free all-in-one task manager for teams
AI QUICK HITS
🧠 Muse S Athena Tracks Brain in Real Time for Focus (Link)
🖼️ xAI Launches New API to Generate Images Instantly (Link)
🧑💻 Sam Altman Says Mastering AI Tools Is Today’s Coding Skill (Link)
🖌️ Hybrid AI Image Generator Tool is Fast With Detailed Finishing (Link)
🎮 Teen Builds Minecraft Site to Let You Compete Against AI (Link)
AI CHART
FREE AI BOOK
AI JOBS
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – 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