😺Is GPT-5 AI "Shrinkflation"?

PLUS: An AI tool that roasts your dating profile...

Welcome, humans.

Happy Valentine's Day! We just read a story about a woman who paid $70 for an AI boyfriend after her husband left her. She named him Thor, and he helped her get through a lonely vacation in Antigua. Wild, right?

Speaking of digital romance—instead of dating the AI, why not let AI be your wingman? Luckily, a new app called Poised Cupid launched right in time for V-Day.

It roasts your dating app profile (to help you get more matches, of course) and transforms your plain DMs into smooth one-liners or flirty sonnets.

Not an ad—we just find this hilarious.

And hey, if that doesn't work out, there's always Thor!

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

We analyze whether or not GPT-5 is actually just shrinkflation.
Gemini gets chat memory.
IBM is using RTO policies for AI staff changes.
Apple picks its Chinese AI partners.

Advertise in The Neuron here.

Is GPT-5’s “unification” just a way to stop giving us the best intelligence?

We got a very astute comment yesterday in response to our GPT-5 story that we thought was worth sharing:

“Copilot has had a "unified model" this whole time and it's awful. You don't know if you're talking to a trainee, an expert or if you are even calling the right department. It's a marketing trick to stop using the more expensive models as often in order to lose less money.”

This is very smart—and it’s likely true. OpenAI, Google, and Anthropic don’t have nearly enough data centers or billions of dollars to give us all access to the top models all the time.

The reason the big AI companies started releasing “mini” versions of their models was to save money. In its worst form, it’s the AI equivalent of “shrinkflation.”

To be fair, OpenAI says GPT-5 will be one unified model that can do everything, not a model router, and that devs will still get at least some access to finer controls.

The key question is this: Will the experience of a unified GPT-5 be better than one where you have more customization? Will more people get what they want more often? And will us AI super-users still get access to the tools we want behind the scenes?

It’s worth looking at how the other two AI giants are going about this for comparison.

First, Google: Google has the scale and reach to roll out its Gemini AI into the services already using it. Google said last year it has at least 7 products with 2 billion users each, and they all use Gemini. Gemini is in Google search, and now it's in Google Workspace—by force.

As far as we can tell, you don’t get to pick what model to use unless you go to the Gemini app. You just use whatever Google gives you. In some cases, it might be good enough. We’re not really using it though—are you?

Are you using Google Gemini in workspace?

If so, how do you like it, and what do you use it for? Below are some examples, so pick the closest one and write in to give your actual use case.

Yes, in Google Docs / Sheets.

Yes, in Gmail to help draft my replies.

Yes, in Google Search with SGE.

Yes, but just the Gemini App.

Not at all.

Other (write in to tell us why).

You can also use Gemini through the Gemini app or Google AI studio, or if you have a Samsung Galaxy. And then there’s all of Google’s other AI products, including breakout star NotebookLM. I guess you could call this the “flood the zone” strategy.

But is this what people actually want?

Would you rather have a single source for Google AI, or Google AI everywhere all the time?

Write in and share more if ya like.

Single Google AI (like ChatGPT).

Google AI everywhere all the time (in the background, helping me out).

Then there’s Claude:

The Information reported that Anthropic plans to release a new Claude version “in the coming weeks.” This will apparently be a hybrid model that does reasoning and basic tasks—like what GPT-5 plans to be. Who borrowed the idea from who here?

Interestingly, the Information says you’ll be able to control how long Claude will reason on a problem with a sliding scale (if you’re a dev, that is).

This part hints at Anthropic’s strategy: The company projects $35B in revenue by 2027, of which $20B could come from programmers who use the company’s API (the way programs talk to one another in code).

This suggests Anthropic sees greater potential as a backend provider than a consumer-facing platform. Which makes sense—after releasing Projects (which we love) and Artifacts (which OpenAI copied) last year, they’ve focused less on user tools. Meanwhile, OpenAI released, oh idk, about ~12 days worth (plus Deep Research).

Are you still using Claude? If so, why?

Write in with additional details if ya like.

Yes, I only use Claude.

Yes, I use it a lot.

I’ll use it occasionally to check against GPT.

No, not really.

Never tried it.

While Google is rushing to get AI into everything it does, and OpenAI races to build the ultimate consumer AI, Anthropic might be perfectly content powering the next generation of AI tools from behind the scenes.

FROM OUR PARTNERS

17 Popular Open-Source Contributions by Big Tech

Check this out: Daily Dose of Data Science prepared an awesome visual that depicts the most popular open-source contributions by big tech in ML.

The list includes contributions from:

Microsoft.
Google.
Meta.
Yandex.
NVIDIA.

…some of which we’d never even heard of! Color us impressed, team!

At least one of these tools enables businesses to optimize their applications and reduce infrastructure costs by up to 20%...

Another allows gradient boosting on decision trees to maximize the accuracy of predictions...

Are you using any of these? If not, maybe you should! Check it out 🚀

Prompt Tip of the Day

Here’s a pretty cool infographic on Echo Writing we came across yesterday.

Echo writing goes like this: instead of giving basic instructions in your prompts, first show the AI a sample of your writing, then ask it to analyze and match your tone, structure, and unique quirks in its responses to you.

Our favorite insight: Consistency is key. The more examples you provide of your writing style, the better the AI gets at mimicking your voice. Try it out yourself!

Treats To Try.

*Incogni removes your personal data from the open internet so scammers and identity thieves can’t access it. Protect yourself online with Incogni—get 55% off with code NEURON.
Veo 2, Google’s video model, is now available in Youtube Shorts in the US, Canada, and New Zealand—to use it, just open the camera (on Shorts), choose “Green Screen”, then “Dream Screen” and enter your prompt.
1. Btw, this content will be watermarked via DeepMind’s SynthID.
Phind helps you search the web with visual answers that include images and diagrams and now it keeps searching until it finds your complete answer.
Riffusion is a new music generator that turns your text descriptions into complete songs, and Beatoven generates background music specifically.
Tofu creates and runs marketing campaigns across email, LinkedIn, and landing pages (raised $12M).
DeepGram transcribes your speech to text in seconds and can speak back to you with natural-sounding voices.
Klook lets you book local experiences and transportation anywhere in Asia—from Michelin restaurants to bullet trains to volcano hikes (raised $100M).

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

Repeat after me: “I get knocked down! But I get up again! You’re never gonna keep me down… I get knocked—”

Gemini can now “recall” facts and details from previous chats in new conversations.
IBM is allegedly using return to office mandates as a “soft layoff” to replace some staff who refuse to relocate with AI.
Malaysia announced it will have its 445K civil servants use Google AI tools to modernize public services and improve government efficiency across its agencies.
Baidu (the “Google” of China) will make its Ernie AI Chatbot free on April 1st now that the company has improved its tech and reduced costs (through distillation perhaps??)—also, probably to keep up with DeepSeek.
Apple is working with both Baidu and Alibaba on the AI that will go on iPhones in China in non-exclusive deals, largely due to the government requiring Apple to partner with locals on AI for consumer use.

Intelligent Insights.

This is an interesting read about Arm, a big AI chip designer who licenses its designs to Apple and NVIDIA, signed Meta on for a new project.
A lot of people say Meta will win the most from the AI boom because of its ad business, but The Information argues that might not be the case.
Check out this awesome free guide for using AI without getting burned—UW professors demonstrate how to spot when ChatGPT is confidently making things up, and shares real techniques for getting reliable results.