Welcome, humans.

Here’s the thing: ChatGPT’s image generator is good… and its ability to create text in images is good… but for whatever reason, its ability to properly label European countries, is, well, not so good:

If it makes you Europeans feel any better, ChatGPT messed up a couple US states, too. Shout out to Nordaka, “South”, Splorida, and my personal home state of “NO.”

Maybe now we know why OpenAI always takes forever to launch new AI models in the EU… it didn’t train enough on Europeans’ data because of those pesky regulations, and now it’s just embarrassingly bad at Euro-geography!

Meanwhile, Meta just announced it will start training its AI on Europeans' public IG and FB posts. Why? To find “FRAFT” on a map, of course.

Here’s what you need to know about AI today:

• OpenAI released a new model for coders.

• NVIDIA started US AI chip production.

• Hugging Face bought robotics company Pollen.

• Apple unveiled plans for on-device AI training.

OpenAI is doing a whole week of launching new things… here’s Day 1:

OpenAI kicked off the week with a bang, but maybe not the one everyone expected. Instead of a new ChatGPT for the masses, they rolled out a new family of AI models—GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano—but strictly for the developers (for now).

Rant: GPT “Four point ONE?!” Aren’t we already on 4.5?!? Pretty sure they're just picking random decimals out of a hat at this point—next week we'll get GPT-4.π followed by GPT-4.½ cup sugar, bake at 350°. Oh well. At least they know.

Here’s the details (livestream video):

• Coding Champs: GPT-4.1 is reportedly way better at coding tasks. Developers are already noticing improvements in their workflows.

• Better Listener: It's significantly improved at actually following instructions, especially complex ones.

• Long Memory Lane: These models can handle a whopping 1M tokens of context and are better at understanding information buried deep within long documents or conversations.

• Faster & Cheaper: OpenAI is touting these as faster and more cost-effective, with the tiny GPT-4.1 nano being their cheapest API model yet ($0.12 per 1M).

That last part is key—OpenAI folks are saying you can “do 4o-quality things 25x cheaper with 4.1.”

Now, this dev-focused launch means you won't see “GPT-4.1” in your ChatGPT dropdown, though. OpenAI says many of its improvements are being gradually baked into the GPT-4o you use daily.

If you do want to try it out as a normie, you can test it in the OpenAI playground (which is like AI Studio, but for ChatGPT instead of Google).

Meanwhile, Memory might be the REAL upgrade: While not today's news, it's worth remembering (pun intended) that ChatGPT's Memory feature got a significant upgrade last week.

The big change? It now learns from your entire chat history, alongside the facts you explicitly tell it to save.

Think of it like this: you can still nail down key details directly (“Remember my company uses AP style”), but it also picks up on your preferences from past chats (like noticing you always ask for bullet points).

This has some solid potential, especially for work:

• Less Repetition: Stop re-explaining project specs, client details, or company style guides.

• Better Consistency: Helps maintain brand voice for marketers or stick to coding standards for devs.

• Faster Workflows: Quickly recall info needed for sales follow-ups or project status updates.

In theory, this update lets ChatGPT move from a generic tool to a more specialized assistant that understands your specific work context. Just remember to manage what it remembers, and be mindful of sharing sensitive data.

To turn it on, go Settings > Personalization > Memory > “on.”

Our take: OpenAI is clearly focusing on releasing these new dev-friendly models to launch its A-SWE coding agent (our guess? sometime later this week).

Meanwhile, the average user gets trickle-down improvements via GPT-4o. You’ll also see the benefits in the wave of new apps built with 4.1—but that’ll take some time.

Last thing: The Information reported that OpenAI's upcoming o3 and o4 AI models will reportedly be used to power its new $20K a month scientist agent, which promises research “innovation” by its ability to connect scientific concepts from different domains. Apparently, multiple Argonne National Laboratory scientists were genuinely impressed during testing.

But who’s actually going to pay $20K a month for that thing? Maybe if it was $20K a year. Imagine paying a $20K salary for a senior researcher. Now THAT would be compelling.

This $20K a month agent got us wondering…

FROM OUR PARTNERS

💼 Want to become an AI Consultant? 

This stat is WILD: The AI consulting market is projected to explode from $6.9 billion to $54.7 billion by 2032. Want to capture this opportunity?

So how do you, an AI enthusiast, become a handsomely compensated AI consultant?

Well, The AI Consultancy Project from Innovating with AI has already helped 300+ students build successful consulting businesses. 

Current students love:

• Tools and frameworks to find and serve clients.

• 6-month roadmap to building a 6-figure business.

• Methods to land your first client in as little as 3 days.

As a Neuron reader, you get early access to the next enrollment cycle!

Take advantage of the AI boom—request early access to The AI Consultancy Project ASAP!

Prompt Tip of the Day

If you’re using ChatGPT with memory, here’s a prompt from this reddit thread to turn GPT into an AI coach to analyze your past convos and help you find your blind spots.

Treats To Try.

1. Whatting transforms your iPad into a customizable diary where you combine widgets for planning, journaling, and tracking your day.

2. Freepik editor lets you cut, stitch, and export videos directly within your workspace.

3. SpreadSimple turns your Google Sheets into a complete website in minutes; just update your spreadsheet and your site updates instantly—free to try.

4. Imagerr creates SEO-friendly image alt text—paid only ($19-$99 one-time purchase).

5. Luna tracks your product goals against delivery data, flags risks before they derail projects, and creates status updates automatically from your existing tool —free to try.

6. Submarine Player creates subtitles for any video you watch on your computer—free with limited features, or $19.99/year for unlimited (mac only rn).

7. Nily puts 20+ models in your browser sidebar, letting you compare results from each as you do all your marketing tasks in one place—free trial, then $19/month.

See our top 51 AI Tools for Business here!

Around the Horn.

• NVIDIA announced plans to manufacture up to $500B worth of AI infrastructure in the United States over the next four years.

• Alibaba’s AI Quark surpassed ByteDance’s Doubao as China’s most popular AI app—Alibaba also previewed FantasyTalking, which turns any static portrait into lifelike talking videos with precise lip-syncing and natural body movements.

• Hugging Face acquired French startup Pollen Robotics, marking its entry into robotics with the open-source humanoid robot Reachy 2.

• Google created a new model called DolphinGemma AI to decode dolphin communication.

• Apple announced a new privacy-focused AI training approach that analyzes user data directly on devices to improve its AI systems.

• Adobe just invested an undisclosed amount into AI avatar company Synthesia.

FROM OUR PARTNERS

Speak to anyone, anywhere–instantly

Solaria is the fastest and most accurate speech-to-text model offering native-level accuracy in 100+ languages, including 42 that are unique to Solaria.

So whether you’re building industry-specific voice agents or delivering high-performance customer experiences, Solaria provides the foundation to expand into untapped global markets.

Try Solaria for free

Tuesday Ticker

Yesterday we asked what your most used AI model was, and the vast majority (almost 400) said ChatGPT 4o. TBH, these days we only use 4o for one off tasks like image generating or SearchGPT—otherwise we use o3-mini high or o1-pro for Deep Research.

Here’s what you said:

• H.P chose GPT-4o: “I have too much context in the one I started in to move over to a new model / platform.”

• S.H. chose Claude 3.7: “Hard choice, depends on the task. I also use ChatGPT o1/o3 a lot (strategic tasks and deep research) and Gemini 2.5 (also ChatGPT 4.5) as alternatives to Claude 3.7 and sometimes switch to Grok 3 and DeepSeek if I need a fresh view on things or an uncommon approach.”

• S.C. chose Gemini 2.5: “Gemini all the way for summarising YouTube and web sites. Great for code (as is Claude). Deep research is fantastic.”

A Cat's Commentary.