- The Neuron
- Posts
- 😺 Your prompts are broken
😺 Your prompts are broken
PLUS: First robot that can actually do your laundry...
Welcome, humans.
Finally, there's a robot that can actually handle that mountain of laundry sitting in your hamper:
Physical Intelligence (PI) just unveiled π0 (pi-zero), a generalist robot model that can unload a dryer, carry clothes to a table, and fold them into a neat stack—without humans.
What makes this special isn't just the folding—it's that π0 can recover when things go wrong, like if your cat decides fresh laundry makes the perfect nap spot mid-fold.
That's because PI trained their model on the largest robot dataset ever, using 8 different robots doing all kinds of tasks.
There’s so many awesome videos of these bots in action, you gotta check them out. Watching π0 ace filling up an egg carton without cracking any eggs is money.
Here’s what you need to know about AI today:
We share when (and when not) to prompt.
Court throws out publishers' OpenAI copyright suit.
FrontierMath tests AI models with expert-level math problems.
GenAI writing found in thousands of fake science papers.
When AI needs the perfect prompt (and when it doesn’t)
Ever rephrase the same question to ChatGPT five different ways, hoping for a better answer? A new study just revealed when this helps—and when you're wasting your time.
The researchers used a framework called ProSA to test a wide range of models (from 7B to 72B parameters) across everything from basic questions to complex coding tasks.
Here’s what they found:
For knowledge-based questions, higher confidence in the subject matter correlated with more consistent answers across different prompts.
As AI models get bigger, they're getting better at understanding what we want—regardless of how we ask.
The research also found that adding just one example output dramatically improved consistency on benchmark tests.
Here are the tasks where AI models were most consistent, regardless of how you phrased the question:
Business solutions.
Expert technical guidance.
IT problem-solving.
And here's where prompting made the biggest difference:
Technical coding and data visualization.
Programming and scripting tasks.
Algorithmic problem-solving.
Here’s the takeaway: you can officially stop obsessing over prompt engineering for everyday questions with big models like GPT. Instead of memorizing prompt templates or magic phrases, you can focus on simply asking for what you want in plain language.
Save your energy for the tasks where precise prompting still matters, like creative work or specialized technical requests. If you're building AI products or tackling complex technical tasks, prompting still matters—a lot.
Pro tip: Gauge the AI's confidence on every subject matter. When models respond with high confidence, they're remarkably consistent regardless of how you phrase things. Lower confidence? That's your cue to try rephrasing.
FROM OUR PARTNERS
Gain the AI Advantage in Business and in Your Career…
The New M.S. in Artificial Intelligence in Business (STEM), offered by Fordham University’s Gabelli School of Business, readies you for the future of business.
Choose from two tracks—Finance Industry or Technical. Flexible schedules for working professionals.
Harness the power of AI to:
Improve Decision-Making.
Enhance Customer Experience and Service.
Streamline Operational Efficiency.
Increase Product Innovation.
Deepen Competitive Intelligence.
Improve Marketing ROI.
Acquire the knowledge and skills that will set you apart. Seats are Filling…
Apply Today for Spring or Fall 2025
Treats To Try.
*Join AI experts from leading companies like Moderna and S&P Global on Nov. 14 at Section’s AI: ROI Conference—a virtual event for leaders looking to achieve tangible results with AI. You’ll hear predictions from Scott Galloway AND real AI ROI case studies. Register for FREE here.
Diaflow helps you build workflow automation apps with AI.
CalendarPlusAI breaks down your goals into daily tasks and builds you a personalized schedule (early stage project).
Elv.ai filters toxic comments and automates social media replies while tracking audience sentiment.
Genbler combines seven unique visual AI tools in one platform.
Voila lets you write emails, research, and create content from any webpage via built-in prompts directly in your browser.
Sincerely Karen turns your bad product experiences into into professionally-worded complaints.
*This is sponsored content. Advertise in The Neuron here.
FROM OUR PARTNERS
Need an AI notetaker for your sales calls?
We like using Attention because it helps us find relevant data from our calls.
Then, once our meetings are done, we can go back and watch the relevant sections, and even schedule automatic follow-ups with Attention’s AI agents.
Around the Horn.
OpenAI scored a quick legal win when a judge threw out a copyright lawsuit against it from two publishers who couldn’t prove a “concrete injury.”
There's a new AI model benchmark called FrontierMath that’s filled with expert-level math problems that take specialists days to complete.
Science mags like Nature are raising the alarm now that GenAI can create ultra-realistic scientific data; for example, LLMs wrote about 7-17% of ~50K computer science articles published between 2023 and 2024.
Monday Meme
A Cat's Commentary.
That’s all for today, for more AI treats, check out our website. The best way to support us is by checking out our sponsors—today’s are Fordham, Section School, and Attention. See you cool cats on Twitter: @noahedelman02 |
|