If you finally finished creating GHibli, listen to me.
The Openai trench grows almost frightening.
Two months ago, in The second edition of this newsletterI said you will pay $ 20 a month for “GPT Co -Coys”, which in turn calls LLMS, which is likely from different companies, in their products.
It turns out that Sam Altman is building some of the best of these covers there.
These are some things that Openai publicly launched in recent weeks:
-
The original images generation in GPT-4O
-
Long -term memory in Chatgpt
-
O3 and O4-MINI models
-
GPT 4.1 in the application programming interface, directed towards coding (and cutting off the GPT 4.5)
-
The coding factor on the “manuscript” device
The 4.1, O3 and O4 models are all very impressive, and I am talking about it in a later section, but these models make Openai a unique position.
In fact, Claude 3.7 of its competition from the competition, Gemini-2.5-PRO of Google, presented a great value similar value, but it fell in terms of attracting the attention of the wider world.
It is the other “smaller bits” on the margin that seems to give an indomitable edge.
Keep memory
Openai has increased significantly from the Chatgpt capacity to remember the information it has told over for a long time.
There are not enough people talking about how important this small step is.
Chatgpt is innocently tries to put herself as your friend, Your best friend. Its tone becomes more and more friendly and is better compatible with your expectations over time.
This makes you share everything with her – what you look, and what your family and friends seem to (Yay GHibli), your travel plans, the gym routine, and what you eat.
We forget that, you are telling him everything you want to do in life, deeper insecurity, and aspirations.
I know at least I do. This is a controlled version of what you already know about me.
It knows my deepest aspirations and a few of my insecurity already cases (I do not participate in the last thing).
This Chatgpt Memory Memory gives a very special feature when you think about it with the early engine feature.
Chatgpt can continue to customize responses, and this would make the user find it “better” than any newer option. Imagine this more than five and ten years. Openai will be known You, In all your glory.
I don’t know what I feel about this so far. I can use GPTS locally on my system or use completely open source alternatives, but being a good and useful product is very attractive.
But just imagine if Mark Zuckerberg always has his ear on our activity on his platforms, allowing him to provide such special ads, how Sammy will be able to expand in our lives with this unprecedented personal information.
Meanwhile, try this claim and see what you get:
“Based on everything you know about me, and the deepest aspirations and insecurity, I tell me 3 things that I must do now after I think I am completely missing them.”
I am surprised by my results.
Openai water test on the coding agent?
Many people may not remember it now, but Openai used to present a “coding agent” via API Codex again in 2021.
Distilately, stop overall access to the CODEX applications interface. Instead, it focuses on the GitHub Copilot app with its main investor, Microsoft.
But GitHub Copilot cannot simply find its success with users, and fail to get the market share compared to columns such as the indicator, the beloved, and the bolts.
At the same time, the market continues to get more profit, as it merges more and more Devs AI into the workflow.
It seems that Openai may not want Github Copilot or Microsoft to hinder this market, and the company has revived the “Constitution” this time as an open source Cli!
I played with Codex for some time, even making Tic-Toe-Toe game!
It is somewhat appropriate to do this task, and any company or company can technically conclude in the user interface.
For me, any Openai’s models are still compared to the Sonnet line of the Claude or Gemini-2.5-PRO from Google in the quality and pricing set, so it will not become my choice anytime soon!
However, it will be interesting to see how easy Openai is in this path of the agency agents through cases of use.
There is a little imbalance between the application layer products such as the indicator (which wants to reduce the number of codes consumed for each query) and the founding AI companies (which benefit from more symbols that are consumed).
This conflict led to problems, as with SonNet 3.7 and Gemini-2.5-PRO in the index. Therefore, it will be interesting to see how Llm founders brings their hands here different results.
Fix the large pricing of Openai
You may remember to remember several times that Openai cannot get a customer of applications programming programs due to prices, with Google and Deepseek like these sweet deals around them.
Openai has finally absorbed the comments that were largely uniform in the DeV world and issued some of the most affordable models on the market, including a new major model of coding in GPT-4.1.
It also suspended the controversial GPT-4.5-PREVIEW model and is charged with applications programming facades.
The company released the O3 model for the public; It is now a higher thinking model that can also integrate images.
He wants to start stealing jobs worth $ 60 trillion
An emerging mechanical is supported, saying that PLBS is paid more than 60 trillion dollars annually all over the world to do a portfolio, and you want to replace them all with artificial intelligence factors.
This advertisement is so much that I was sure that it was a mockery at first, but it doesn’t seem so.
The startup is managed by former employees of the Amnesty International Research Foundation, EPOCH.
Other events
- Illon Musk launched the Grok Studio, following the industry direction of allowing users to “cooperate” with artificial intelligence more, rather than just bringing responses in one direction. It is very similar to ChatGPT and CLADE masterpieces in my experience.
- Openai may cost “tens of millions” of dollars by being polite to chat and say thanks to you, according to Samtman (I think this was just a Simon note).
- Instagram is supposed to try to capture teenagers who lie about their age and restrict their accounts, using artificial intelligence (via Techcrunch).
- Figma makes “Applications” of artificial intelligence based on the LLMS. Hof, do we really need more of these? Better to find the benefit of Amnesty International in improving the functioning functions of the UX user interface.
- China offers an Amnesty International Curriculum to School Students (via Reuters).
It is better from all over the web
This Redditor went to 7-8 physiotherapists over the years, but he was unable to find a solution for back pain-even Chatgpt!
Each physics went to give me one person, one angle. But with ChatGPT, I get a summary of all known therapeutic knowledge of the human being, it was liquidated through a personal context more than I can give Vizio no longer an hour, designed according to the specified learning method. Don’t make it look like an advertisement but … the best $ 20/month has ever spent.
U/SoundBoy89 on Reddit
This higher suspension of U/Typo180 also strikes the right home. “One of the magic parts of LLMS is that she never feels bored and has no place elsewhere.”
Mimi the week
This publication of the thirteenth edition is artificially reinforced.
