AI Changed Everything This Week | INCREDIBLE NEWS | The Hi-Tech Discussion

Last week I disrupted the usual flow of news to discuss Tucker Carlson's interview with Vladimir Putin, and analyse the foundations of Putin's persona.

This week we are back in tech, but something else disrupts the usual reporting: We just one through one of the wildest seven days in AI history again. Seven days that are changing the world again.

Let's dive in.

Gemini 1.5

Google DeepMind 's Gemini 1.5, the competitor to ChatGPT, will now be using the Mixture-of-Experts architecture – in easier words, a model comprised of smaller models (experts).

When you write a prompt, the main model decides which of the experts should answer the prompt. This increases efficiency, enabling prompts to be processed with less data use.

The context window is also increased: Gemini 1 had 32K tokens (in words, 24K words of input and output), whereas Gemini 1.5 gives you 1 million tokens – around 750K words of input and output text. To give you an idea, the entire Lord of the Rings saga amounts to 480K words... that's the amount of data this model can now accurately analyze.

OpenAI: Sora Shakes The World, ChatGPT Remembers, and Sam Altman Wants MONEY

If you thought AI generated videos were far from good, Sam Altman and his crew decided to prove you wrong: Sora is capable of generating uncannily realistic videos, and there is even a full research paper about it that I highly recommend you read to see also video examples (find it here).

Sora can also generate images, and you can see one in the example above.

This outstanding model is, in classic OpenAI fashion, disruptive. It is also worrying: It's all fun and games until, next year, you're sitting in court looking at an video of you committing a crime you're not guilty for.

On Twitter, I've seen also remarkable examples of animated characters that could even put Disney out of business. The quality of the Sora output will also become a stable for content creators, putting out of business stock photography and videography – and saving a lot of time in creating B-Roll.

ChatGPT Has Memory

OpenAI is slowly rolling out a new feature: ChatGPT now remember details from your conversations with it, and store it to use it as context future discussions.

Now, this data will be used to train the models, and OpenAI was fair in this regard: By going to Settings -> Personalization, you can both turn off the memory feature, and check which memories ChatGPT is storing, with the possibility of even deleting some or all them.

On top of that, there is also an "incognito" mode called Temporary Chat where your information won't be stored and won't be used to train models.

Cool, right?

Sam Altman Wants the GDP of a Country?

We've seen some reports claiming that Sam Altman would like to raise $7.000.000.000.000 (yes, TRILLIONS) to develop the next stage of AI and finally close in on developing AGI.

Turns out, this is false. What Sam was referring to is the total amount of investments, assets, energy use, and manufacturing capabilities that might be necessary in the coming years to get to AGI.

Stable Cascade: AI Art + Legible Text = Logos?

Stability AI released Stable Cascade, an amazing AI model that can finally generate art that includes legible, mistake-free text 9 times out of 10.

The speed of generation is remarkable. To tinker with it, go to

The quality of the output is honestly outstanding, but I want to clarify something for all the AIficionados here: No matter how good your AI models get, they can't design logos. What people are claiming to be "logos" are just simple drawings.

An AI generated logo is meaningless. A real logo has to come from strategic work – it has to communicate a message to the audience that's both novel and familiar. It has to be thought-out.

For example, your last name, written on paper, is just a word. It means nothing without the pride, the memories and the emotions you link to it. A logo is the same: without the work behind it, you might as well not have one.

So, have fun with Stable Cascade. Got to Hugging Face , type "Stable Cascade" and go to work. But unless you want your company to look like you ran a high school art contest for a logo, please stop generating those.

In Other News

• Zuckerberg Reviews Apple Vision Pro: In full 2024 style, Mark Zuckerberg showed up looking jacked under an oversized t-shirt, in a video that looked recorded on a 2014 Android phone... and he reviewed Apple's new Vision Pro. In a nutshell: Except for the eye-tracking feature, Mark thinks he has the superior product. Unfortunately for brand, he's got the weaker brand as a company... but I'll give it to him, as a person he is becoming way more likeable.
• Make money with your voice: So, let's say your voice sucks but you want to have an audiobook. Or let's say you need voiceover for videos. Well, ElevenLabs has got you covered: You can now upload your voice to their system, and whenever someone uses it you will basically get royalties.
• Chat with RTX: If you have an NVIDIA RTX30-series and above, you can now access models stored locally on your system. The chat, which comes as a 35GB download, will let you analyze documents, links, and even YouTube videos.

⚜️ That's it. If you enjoyed this article, follow me Sebastian D. P. Hidalgo – every week I give you tech news so you don't have to spend time scrolling.

♻️ Also, do share with you audience. Thanks in advance.