- Golden Horizon AI
- Posts
- 🔥Microsoft releases their AI model
🔥Microsoft releases their AI model
ALSO: OpenAI in medical research + AI Army guide part 2
🔥Microsoft releases their AI model
ALSO: OpenAI in medical research + AI Army guide part 2

Estimated Read Time: 5 minutes
Microsoft wants AI on your phones and pc, medical science is using AI to find new solutions, and Meta’s stock is the one to buy right now. Let’s dive in!

🔥Microsoft releases Phi 3 to compete with Llama 3.
😷Moderna uses OpenAI for medical research.
📉Meta’s stock free-falls 15% after market close.
🎥Hollywood is creating AI clones of celebrities.

Read time: 1 minute
🔥Microsoft releases Phi 3 to compete with Llama 3

What happened: Microsoft just released Phi 3, a very small AI model (often called an SLM) that somehow competes with larger models like Llama 3. It’s also optimized to run right on your smartphone.
The details:
The model boasts 3.8 billion datasets, but competes with models over twice its size.
The point of the smaller size is to fit this locally on your machine, like your smartphone.
Microsoft claims the secret was data quality, not quantity. They use fewer but higher quality data sets.
Researchers also introduced a larger variant, the phi-3-medium with 14 billion parameters.
Their performance shows AI models may have reached a plateau from raw data alone and may need more quality.
Technical Specs (for the nerds):
It features 32 heads, 32 layers, and a hidden dimension of 3,072, with a vocabulary size of 320,641.
The model can process over 12 tokens per second on an iPhone 14, requiring only 1.8GB of memory when quantized to 4 bits.
Phi-3-mini is built using a transformer decoder architecture with a context length of 4K, expandable to 128K in its long context variant.
Why it matters: This is the direction of big AI tech. Microsoft, Meta, and even Apple have shown they want AI to work locally on your PC/smartphone. They will likely work in conjunction with larger AI models in the cloud for complicated tasks.

Stock | Previous Close | Closing Price | Change (%) |
---|---|---|---|
Amazon | $179.54 | $176.59 | -1.64% |
Tesla | $144.68 | $162.13 | +12.06% |
Meta | $496.10 | $493.50 | -0.52% |
$158.26 | $159.13 | +0.55% | |
Apple | $166.90 | $169.02 | +1.27% |
Microsoft | $407.57 | $409.06 | +0.37% |
Nvidia | $824.23 | $796.77 | -3.33% |
As of April 24, market close.
(Some serious stuff is happening with Meta though. Read below for more details…)


😎 Ray-Ban and Meta’s new glasses now include AI from Llama 3. You can talk to the glasses and ask them questions about what you’re seeing. Contrast this with humane’s AI Pin, which is getting continuously lambasted for being useless. I think glasses have much more potential. These puppies will run you $300 and up.
🚗 Charges have been filed against Tesla driver for killing a motorcyclist. He claims he trusted Tesla’s self driving function so much that he left all the driving to the AI. Then it hit someone. Side note: Tesla is looking into completely self driving taxis this year. Let’s see how that pans out.

Read time: 1 minute
😷Moderna uses OpenAI for medical research

What happened: AI is getting adopted into medical research. OpenAI just revealed their latest customer story. Moderna, a leading player in mRNA tech (the tech the Covid vaccine was made with) has joined forces with OpenAI. The aim is to optimize Moderna's workflows and drive innovation in mRNA-based medicines.
The details:
The partnership uses OpenAI's ChatGPT Enterprise (the professional tier used by companies) to streamline repetitive tasks.
Tasks include predicting how different mRNA sequences will behave biologically.
They’ve made over 750 custom GPTs tailored to various functions within the company. 😳
The use of AI has enabled Moderna to maintain a lean operational structure, despite plans to introduce 15 more products in the next few years.
They claim ChatGPT helped in legal compliance, policy management, and clinical data analysis.
The partnership started in 2023.
Why it matters: This is the future of companies. AI will give them a massive boost to productivity in all repetitive areas. It’s also the future of employment, seeing as Moderna talks about training their staff to use AI proficiently over the next six months. Learn to use AI systems now. It’ll give you the advantage.
Just as an example: Many legal firms have already started using AI to read and create documents. Often times the documents the AI reads are… also created by AI.
Note: If AI can do all of this and still can’t find a cure to cancer, I’m putting on my tinfoil hat.

Read time: 1 minute
📉Meta’s stock free-falls 15% after market close

There’s just… no soul in those eyes, man.
What happened: All it took was a dose of realism from Zuck himself. Meta's shares dropped over 15% (after hours) following a conservative sales forecast for Q2. And it’s mostly because of lowered AI expectations.
The details:
This followed immediately after Mark Zuckerberg stated it will take years to make a profit from AI.
AI spending has been in the billions for all tech companies, including Meta, with no immediate sign of return.
Meta's Q1 revenue soared by 27% to $35.64 billion, marking its fastest growth since 2021.
Meta let go around 10% of their staff (most tech companies are laying off).
Reality Labs (the virtual reality crap) continues to incur losses, reporting a $3.85 billion loss for the quarter.
Meta is expected to spend even more on AI infrastructure in 2024, again with no return for years.
Some speculate this may be evidence of an AI bubble created from overhyped expectations.
This is also an amazing opportunity to invest in Meta, since they will likely bounce back from this and continue to climb.
Why this matters: What Zuck said is true, AI is playing the long game. Amazon and Microsoft plan to spend tens of billions on AI infrastructures, which won’t pay off for years. This serves as a bucket of cold water to investors who expected these investments to pay off anytime soon.
Pro-tip: Invest in Meta right now while it’s low. It’ll keep going up again.

Read time: 1 minute
🎥Hollywood is creating AI clones of celebrities

What happened: It’s the attack of the clones! A top Hollywood talent agency, CAA, has reportedly been testing an initiative called CAA Vault. Basically it clones celebrities using AI. So I guess we’ll just be watching their likeness now. Yay. 🫠
The details:
CAA is partnering with AI firms to scan every part of the actor’s body.
This data creates a digital reconstruction of the celebrity, which can “act” in their place.
This comes after OpenAI spoke with Hollywood directors about Sora, their video generation AI. It has a lot of industry workers on edge.
It’s not too different from how real life actors are scanned into videogames.
Takeaway: This honestly isn’t surprising, and I’m surprised more agencies haven’t done this. In fact, movies already use digital recreations of actors all the time. Watch any Marvel movie and you’ll witness more cartoon than reality. But this could change actors from… actors and more into “likeness” license holders. The industry is about to go through some upheaval.

Read time: 1 minute
🤖Create an army of AI workers: Part 2

Day 2 of our GPT Guide:
So now you know how to create a custom GPT.
Today we’re going to learn how to use these GPTs for content, specifically with social media.
The lessons you learn today can be applied to any form of content creation, whether that be a book, blog post, or other social networks.
To do this we’ll use the instructions below AND provide training data. Training data are files you upload into the GPT as reference.
This will be specifically for Twitter, but it works for any social media platform. Just change the character limit.
To start, enter these instructions:
Create a social media content assistant for my business.
The tool will be used to generate social media ideas and posts for you my business.
You will use the uploaded social media posts and analytics as knowledge of my past social media content.
Analyze the analytics to identify the most successful content and aim to replicate that content.
Step 1: Use the training data to generate a list of content ideas and ask me to identify the idea to focus on. Identify the main content pillar for each idea.
Step 2: Ask which content level I want the content created for.
The content levels are as follows:
Level 1: Wide scale, wide audience appeal. Awareness generating viral content. Use the internet to research common topics to intertwine my content with. Especially focus on hot topics that people are emotionally invested in, such as hot news topics or celebrities. Focus on viral or controversial news topics. This content must be simple in complexity and simple for a wide audience to understand. Short length posts. Maximum character count is 280.
Level 2: Medium scale, medium audience appeal. The content has medium complexity with medium length. Expert opinions and hot takes on common topics within the niche. Maximum character count is 280.
Level 3: Small niche audience content. High complexity and longer post length. In depth, not for beginners. Long form content. Max character count is 1,500.
Once you know which content level we are working with here, move onto step 3.
Step 3: Create multiple posts about that topic for the selected content level.
Do not include hashtags or emojis
Do not use immature language or repetitive words..
Match my tone of voice and style.
The post provided should be the whole post, not a preview or snippet.
Do not provide the levels in step 1, only in step 2.
Make sure to alter this input for your social media style.
Now that you’ve entered the instructions, click on the usual custom GPT options like the name, image, description, etc.
Now we’re going to upload our files as reference. This is what ChatGPT calls “Knowledge”. Just hit the upload button.
Try to have only one GPT per social media account/type.
For example, in X/Twitter you can download a zip file of your data.
Now select the data and analytics of your specific platform. Upload them to the GPT. Make sure the GPT has all the data it needs, like the posts, the “likes” of each post, views, comments, etc. The more info the better.
Now test out your GPT. Don’t expect things to work flawlessly. Something will go wrong or be imperfect, which is what you want. Adjust the GPT input to fix the problems.
Now here’s something extra.
You can use this GPT to create posts similar to your competitors/inspirations. This is especially useful if you’re still new and finding your “voice”. All you need to do is upload their data to the GPT instead.
Again, I’d recommend creating a separate GPT for this.
This can be used for anything, such as emulating the style/voice of popular blog writers or Twitter posts.
This can even be used for Video scripts, since you can use GPTs like this one to summarize videos. Then you can upload those summaries or scripts.
And you can emulate various parts of each person, such as one person’s voice and another person’s formatting style. Just make sure to tell the GPT which files it should take which type of inspiration from.
More so than anything, remember AI is a tool. Don’t expect any GPT to do literally everything for you. This will likely create fantastic baseline posts, but you should still work to do more with what the custom GPT gives you.
Next time we’ll cover a copywriting GPT. This one is pretty cool what results you can get!

🎮 AI in my videogames!

Looks more like the show but still awesome.

The classic. I miss when he had good games.

He almost looks like a toy.

Yo 🤣
Source: Midjourney

Give us your feedback!
Got feedback? Rate today’s newsletter by clicking below!
Got more personal feedback? Fill in the box when you rate us and give us your criticism and feedback.
Thank you for reading!
❤️Share the (ARTIFICIAL) Love!
Got a friend who wants to appear as smart as you? An AI fanatic or someone shaking in their boots of getting replaced? Share us by clicking the link below and you’ll get our love ;)