OpenAI's New Deep Research Re-Sets The Bar (again)

Cure cancer or win an argument with your spouse for just $200/mo

Author

AI 4Humans
04 Feb

Today on AI For Humans The Newsletter!
This new AI tool just broke all the benchmarks
OpenAI releases a new o3 model (it’s really good)
And a rough week DeepSeek security
Plus, our can’t-miss AI feature of the week!

Welcome back to the AI For Humans Newsletter!

We talked last week on the show about how Sam Altman’s response to DeepSeek R1 was that OpenAI would “pull up some releases” and whoo doggy, did they.

Not only did we get their updated reasoning model o3-mini for free (take that China!) but they dropped something brand new: Deep Research.

Deep Research (unfortunately also the name of Google’s similar Gemini tool) is OpenAI’s new AI Agent that will turn your more significant ChatGPT queries into full blown research reports. Deep Research runs on the full o3 model and is, according to many, many people, returning absolutely spectacular results.

For now you need the $200/month pro plan and I’ll be totally honest, this is the first time I’m really tempted to go in on it. We’re getting to a place where that insane amount of money doesn’t seem so insane any more.

One final note Deep Research note: Sam tweeted the following…

Which, yes of course, yay for the team! But also, even as just as a vibe check, ‘single-digit percentage of all economically viable tasks in the world’ is a phrase to really take to heart. We are on the steady path towards AGI and we’d better start preparing for that.

-Kevin & Gavin

3 Things To Know

A Brand New Model from OpenAI, o3-mini!
We also got o3-mini, which will appear as o3-mini and o3-mini-high in your model selector. The difference is the amount of time is spends noodling on your problem. These models in most cases are as good or better than the o1 model, and particularly excel at math and programming. OpenAI also announced a version of ChatGPT for US gov’t folk, and an agreement with US National Laboratories.

Google’s Doing Phone Things
A couple AI launches for phones - I mean for calling people or places - the legacy phone function. Remember? First, an AI that’ll sit on hold for you. Formerly just a Pixel phone feature, now rolling out to everyone via Google search. Also a new business solution called “Ask For Me”, but it only works on auto repairs shops and nail salons (not a joke).

Trouble In Paradise for DeepSeek [a bald eagle screeches]
App data including chat histories and API keys were uncovered by security researchers “within minutes”. It’s worth clarifying this is problematic for DeepSeek’s app & model hosting, but doesn’t necessarily extend to someone running the open-source model themselves. Also US AI Czar(!) David Sacks pointed to a study that estimates DeepSeek spent over $1B on it’s compute cluster. Community Notes quickly clarified DeepSeek reported only spending $6M on the final training run.

We 💛 This - Runway Frames

AI images today can suffer from a weirdly synthetic polish. It’s the sort of thing that just won’t do if you’re looking for a textured B movie vibe.

Enter: Runway Frames. 4 out of 5 tortured artists agree “this one is giving.”

This page actually shows a great collection of prebuilt styles you can draw from, by simply selecting it when generating images. It won’t be as strong as Ideogram at more detailed prompts, or images with text, but it’s absolutely nailing a really cool look.

Are you a creative or brand looking to go deeper with AI?
Join our community of collaborative creators on the AI4H Discord
Get exclusive access to all things AI4H on our Patreon
If you’re an org, consider booking Kevin & Gavin for your next event!