Happy Sunday! š
We had a super busy week so The Format is rolling out slightly lateā¦
Itāll be worth it though because this week has been a big one.
Other than the HUGE releases by OpenAI⦠weāll be looking at:
How token prices are killing AI products š§āāļø
Grokās unfair advantage⦠an AI bot that strips for you? š¤
The two sides of Zuck ā
How you can fly/walk/drive around your AI world š
Anyway letās jump in!
Introducing GPT-5 and gpt-oss š
We had a double drop from OpenAI this week. GPT-5 and gpt-oss (yes one is upper case and the other lower, I donāt know why).
There has been a lot of public response to both.
Letās start with GPT-5:
Itās definitely an upgrade, and the way they have unified all the models into the āGPT-5ā model is great. But after waiting for over 2 years, people were expecting and āApple momentā.
And well⦠the general consensus is that we didnāt get one.
My two biggest takeaways from the long release video were:
Everyone who is on the free tier will now get GPT-5, for the people who donāt really know/care what model they are using, itās going to feel a whole lot smarter
They briefly mentioned an integration with Gmail and Google Calendar. This feels like a huge thing, and with some more tactical integrations the personal context the models will have, and actions they will be able to take will make them exponentially more useful.
Now onto gpt-oss:
OpenAI also released 2 impressive open source models. The first time theyāve been open in quite a long time.
These models arenāt quite state of the art, but they arenāt too far off, and given the fact that most of the other top open models were built by Chinese companies, this release is a big win for American open source.
The fact that a big name like OpenAI has released these models will give so many enterprise customers the confidence to move towards a privacy focused approach.
Our experiments internally have already begunā¦
Tokens are getting more expensive
A lot of AI companies and AI-based products have fatal flaws with their pricing structure. Itās a race to the bottom, and itās causing a blood bath.
When they came up with their pricing model it was under the following assumptionsā¦
year 1: break even at $20/month
year 2: 90% margins as compute drops 10x
year 3: yacht shopping
Thereās a problem with this though.
Although the margins have dropped exactly as expected, customers are expecting the latest state of the art models.
And these models are actually costing significantly more due to āthinkingā time and the vast amount of additional tokens that are required to achieve this.
Companies arenāt making huge margins, theyāre actually making huge losses on power users, itās just not sustainableā¦
The piece does explore quite a few interesting solutions to this problem though.
My favourite is Replitās approach: complete vertical integration.
By making sure that they have infrastructure covering every layer of the stack, and importantly that they make a good margin at each layer, they are able to essentially use the AI development part of their platform as a āmarketing budgetā.
Plus, as a user the switching costs become much, much higher, since you have to switch the whole stack.
Itās genius⦠and itās working.
The World Becomes Playable
When I say āThink about AIā you almost certainly start picturing a large language model like ChatGPT, then maybe a voice or image model.
I bet your brain didnāt imagine a world model.
Wait⦠what is a world model?
Well this week Googleās Deepmind team released a paper on demoās for Genie 3. A model that can simulate a 720p world for you based off a prompt/image/video, then in real-time you can navigate around it and interact with it. Mad. š¤Æ
The most impressive things about the model in my opinion:
The physics⦠this model acts as if it has a built in real-time 3D physics engine (something Unreal Engine has spent billions on). Everything from the dynamic reflection in water to complex motion, itās wild.
The memory⦠although itās currently only minutes long, itās crazy being able to interact with an object e.g. paint on a wall or open a door, leave, go on a short adventure, come back and the world remains exactly as you left it
The range⦠whether itās exploring the world from the first person perspective of a bird, the third down perspective of a Mars Rover or the top down perspective of an 8-bit strategy game. The model can create almost any world, with any aesthetic and from any angle.
Will GTA 6 be AI generated?
Probably not⦠but GTA 7 almost certainly will be.
Other articles + videos weāve found interesting this weekā¦
Manus is launching āWide Researchā that spins up 100+ agents to scour the web for you ā The concept of running parallel agents at once and then bringing together some/all the results is something that we are only going to see more of over the next months and years. Weāve been experimenting internally⦠š
In Search of Native Business Models ā In many āAI productsā it seems like the line between ātoolā and āworkerā is disappearing. For now usage-based pricing is king, but itās only time before a new (hybrid?) pricing model will rise.
The concept of ābillable hoursā is quickly becoming antiquated.
How a āvibe workingā approach at Genspark tripled ARR growth ā In what they call āvibe workingā, Genspark have implemented an AI-native workforce from the ground up. This is allowing them to move, pivot and grow incredibly fast.
When will āAI-nativeā become the only viable way for teams to work?
Zuckās note on Personal Superintelligence ā Rather than industries replacing human work directly, Zuck discusses how he sees AI as a deeply personal assistant that knows your goals, helps you create, connect, and grow, and empowers billions to shape the future on their own terms.
This seems like a lofty contrast to their current business model with Meta pumping on the news that āvideo screen timeā is up 20% thanks to their curated doom scrolling feeds. We shall see.
But I donāt expect self-actualisation to come via Instagram Reels.
xAI Launches Companions That Can Engage in NSFW Chats ā Premium X users can now chat to anime-styled ācompanionsā. They have strong and unfiltered personalities, are very fun to talk too (if youāre into that), and you can be very intimate with themā¦
Ethically questionable? YES. But the strategic data-acquisition play makes a lot of sense. People will talk to it in a way they wouldnāt with Grok.
Means and Meaning ā Whilst technology may be rapidly giving us more means, it doesnāt help with the problem of meaning. This essay by Packy McCormick dives into what happens as humans are abstracted away from productive work and how this is/will impact attention, meaning and life.
OpenAI validates the open ecosystem (finally) ā OpenAI seem to be playing strategic games yet again. They have finally released their long anticipated open source model. But why? How does it benefit them?
One take is that they have released this model to undercut their competitors (Google, Anthropic etc.) and then they are aiming to frame GPT-5 as the state of the art go to model for everything else.
The battle for AI market share continues⦠š
What weāve been up toā¦
Weāve had a little design overhaul. šØ
Out with the black and yellow dark mode. In with purple and light mode.
So far feedback has been very positive!!
I see we have new UI refresh š just dropped in to say I love it
The new dashboard looks amazing in open format! š„
This is just the start though, thereās a much bigger design and brand refresh in the works⦠keep your eyes peeled. š
Andy spent some time walking through the dashboard šŖ
Below he shows some of the questions you can ask your Community Assistant, walks through a sentiment report and sends some rewards š
We š Pods
Internally weāve slightly evolved how we work, and weāre now using a system where we have multiple pods.
Without diving into it too deeply what this should mean is that youāre going to see very rapid iterations of the product and assistant (at least 1 per week).
Weāre super excited!!
Weāve written a CommunityOS Playbook. š§
If youāre interested and want access just let us know!
And thatās all for this week.
Have a great weekend, Dan and the Open Format team šš½
A couple of asksā¦
If you know a community that would be interested in getting a sentiment report for their Discord or Telegram just drop us a message. Weāll get it to you in no time. š¤
Check out our website (feedback much appreciated!!)
Join our Discord to get all the real-time updates on what is being built with Open Format