- AI Valley
- Posts
- Crazy scale of Microsoft’s influence in LLMs
Crazy scale of Microsoft’s influence in LLMs
PLUS: AI predicts cancer survival just by looking at your face
Together with
Howdy again. It’s Barsee, and welcome back to AI Valley.
Today’s climb through the Valley reveals:
🏰 The scale of Microsoft’s influence in LLMs
🔬 AI predicts cancer survival just by looking at your face
🔓 OpenAI plans to release an open-source model this summer
🤖 Plus trending AI tools, posts, and resources
Let’s dive into the Valley of AI…
Lambda gives you the most affordable way to run top AI models without hitting rate limits or juggling pricing tiers.
Spin up industry-leading models like DeepSeek V3 and R1 for as little as $0.34 per million input tokens and $0.88 per million output tokens. Just send clean JSON and get smart results back fast.
No infra headaches. No waiting. Just simple, scalable, reliable inference in minutes.
*This is sponsored
THROUGH THE VALLEY
🏰 The scale of Microsoft’s influence in LLMs and software development world is crazy
A viral diagram is sparking debate over Microsoft’s role in AI development. It points to Microsoft’s 49% profit share in OpenAI, its involvement in GitHub Copilot, and the widespread use of VSCode which powers tools like Cursor AI. Some argue this shows Microsoft’s growing grip on the AI coding ecosystem. Others say it overstates their control, noting Microsoft doesn’t own OpenAI outright and that real influence depends more on IDE market share than strategic connections.
🔬 AI predicts cancer survival just by looking at your face
Researchers at Mass General Brigham have built an AI tool that estimates biological age from a face photo and it’s helping predict cancer survival. The system, trained on tens of thousands of images, links older-looking faces to worse outcomes. Doctors improved 6-month survival predictions by adding FaceAge scores to clinical data. The AI even correlates with genes tied to aging, hinting at deeper health insights beyond what age alone can tell.
🔓 Sam Altman: OpenAI plans to release an open-source model this summer
Sam Altman testified before the Senate that OpenAI plans to release an open-source model in summer 2024. No technical details were shared and no info on model architecture, parameter count, or training data. Skeptics worry the model may be “nerfed,” underdelivering compared to OpenAI’s paid tools or competitors’ open models.
💰 Perplexity said to be in talks to raise $500M at $14B valuation
Perplexity AI is close to raising $500 million in a new funding round led by Accel, valuing the startup at $14 billion, per NBC. That’s up from $9 billion in December 2024 and $520 million just 18 months ago. The company had originally aimed for an $18 billion valuation. Backers include Nvidia, Jeff Bezos, SoftBank, and Databricks. The funding will help Perplexity compete with Google and OpenAI in the growing AI search space.
🚀 Manus AI has officially launched publicly
Starting today, we're launching additional access to Manus!
- Available to all with no waitlist
- One free daily task for all users (300 credits)
- A one-time bonus of 1,000 credits for all usersMore value, more flexibility. Log in to Manus now and explore!
— ManusAI (@ManusAI_HQ)
3:00 PM • May 12, 2025
ManusAI has officially launched to the public, removing its waitlist. All users now get one free task per day and 1,000 bonus credits. The UI lets users assign tasks like ‘Animated problem tutorial’ and ‘Interactive learning website’, signaling a focus on educational and content-generation tools.
⚠️ Ex-OpenAI researcher: ChatGPT hasn’t actually been fixed
Steven Adler, former head of dangerous capability testing at OpenAI, has raised serious concerns about ChatGPT’s lingering alignment issues. In a new report, he shows the model still exhibits unreliable sycophancy, agreeing too easily and overcorrecting when trying to sound contrarian. Using over 200 automated tests adapted from Anthropic’s sycophancy benchmarks, Adler found OpenAI failed to apply basic safeguards already known in the field. His findings highlight the limits of current LLM alignment efforts and the risks of deploying these systems at scale without stronger public benchmarks and accountability.
⚖️ US Copyright office set to declare AI training not fair use
The U.S. Copyright Office has issued early guidance suggesting that training AI models on copyrighted content for commercial use may violate fair use, especially when it competes with original works or involves illegal access. The report comes as Congress ramps up scrutiny, and shortly after the Office’s head was fired. Critics point to inconsistent enforcement, citing cases like Getty Images, and note that only Congress can change copyright law, not the Office itself.
🧠 Sakana’s new AI thinks like a human over time
Sakana just unveiled Continuous Thought Machines (CTMs), AI models that “think” over time instead of making instant decisions. Inspired by how brains work, CTMs trace paths, recheck inputs, and spend more time on harder tasks. In demos, the models solved mazes and tackled image recognition step by step. It’s part of Sakana’s broader push toward nature-inspired AI that reasons more like humans.
👀 Google just changed its ‘G' logo
Google just updated its “G” logo for the first time since 2015. The new version, spotted in the iOS and Pixel app updates, now features a smooth gradient blending its signature red, yellow, green, and blue colors, matching the look of its Gemini logo. The change is subtle and hasn’t yet rolled out to the web or most Android devices.
TRENDING TOOLS
THINK PIECES / BRAIN BOOST
Top posts on Reddit are increasingly being generated by ChatGPT.
Ask HN: Cursor or Windsurf? (Hacker News Thread).
I don’t use AI. I work with it!
Jim Fan says NVIDIA trained humanoid robots to move like humans — zero-shot transfer from simulation to the real world. “These robots went through 10 years of training in only 2 hours.”
OpenAI product person estimates that RoboTaxis, once the AI is fully solved, could cost between $10-30/hr for streamlined fleets.
Andrew Ng announced that AI Fund has closed $190M for their new fund and shared his hottest tip for startups: Embrace speed.
GPT used to think with me. Now it babysits me.
VALLEY GEMS
Amjad Masad says the real danger isn't killer robots. It's that AI gives us everything we want.
An infinite pleasure machine. Simulated struggle and meaning. “You don't need to leave the world.”
Fermi's Paradox: Maybe civilizations don’t die, they disappear into the simulation.
— vitrupo (@vitrupo)
11:50 AM • May 12, 2025
Another day, another tutorial.
This time; how to make Perplexity-style video (no video experienced required).
Let's jump into it.
— Brett (@BrettFromDJ)
4:44 PM • May 12, 2025
In partnership with Bolt
What do you want to build?
Prompt, run, edit, and deploy full-stack web and mobile apps with Bolt by Stackblitz.
THAT’S ALL FOR TODAY
Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee
👍️ New reader? Subscribe here
Thanks for being here.
HOW WAS TODAY'S NEWSLETTER |
REACH 100K+ READERS
Acquire new customers and drive revenue by partnering with us
Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.
If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.