Mira Murati launches "Thinking Machines Lab"
Good morning from San Francisco,
Today I'm leading with a refreshing story about an OpenAI veteran who proves that not all tech visionaries dream of world domination.
Story 1: Mira Murati launches "Thinking Machines Lab"
Former OpenAI strategist Mira Murati has apparently had enough of the "AI will replace us all" rhetoric. With her new startup "Thinking Machines Lab," she aims to demonstrate that AI and humans can be a dream team. Revolutionary? In Silicon Valley, absolutely.
Story 2: X's $44B Full Circle
Remember when everyone thought Elon Musk overpaid for Twitter? Well, X is now seeking new investment at exactly that price - $44 billion.
Story 3: Meta plans LlamaCon
Mark Zuckerberg is hosting the first LlamaCon - a developer conference focused exclusively on generative AI. With an AI budget of $80 billion for 2024, Zuckerberg shows he's ready to buy all the llamas he can get.
Have a great start to your day,
Marcus Schuler
AI Pioneer Mira Murati Launches Human-Focused Startup Thinking Machines Lab
In a move that has Silicon Valley buzzing (and a few robots nervously checking their resumes), former OpenAI CTO Mira Murati has unveiled Thinking Machines Lab. The startup aims to develop AI that collaborates with humans instead of replacing them. What a concept.
The 30-person team includes some heavy hitters from the AI world. Former OpenAI VP Barret Zoph is CTO, and ChatGPT inventor John Schulman has signed on as chief scientist. They clearly got tired of training AI to write poetry and decided to make it actually useful.
Their San Francisco office is already humming with activity, developing AI models that optimize human-AI teamwork. Unlike other AI companies racing to create the next digital oracle, Thinking Machines Lab is taking a more grounded approach. They're even promising to share their work through technical papers and code releases. Transparency in Silicon Valley – now that's truly artificial intelligence.
The company's name pays tribute to Danny Hillis's 1980s AI venture, which pioneered parallel computing before declaring bankruptcy in 1994. Murati's team is hoping their human-centric approach will lead to a different ending. Though they probably won't turn down any profitable side gigs teaching AI to write haikus.
Recent industry developments suggest they might be onto something. While companies like DeepSeek claim they can build advanced AI models at bargain prices, Thinking Machines Lab is betting that the real value lies in making AI work better with humans. After all, what good is a brilliant AI if it can't understand why humans still enjoy watching cat videos?
Why this matters:
- In an industry obsessed with creating artificial general intelligence, Thinking Machines Lab's focus on human-AI collaboration is like finding a vegetarian at a steakhouse – unexpected but refreshingly sensible
- By making AI more accessible and transparent, they could help bridge the gap between Silicon Valley's AI dreams and the average person's need to just get their spreadsheets done
- The impressive team suggests that even AI experts are starting to realize that maybe, just maybe, humans aren't completely obsolete yet
Read on, my dear:
From Twitter Chaos to X Success: A Turnaround Story?
X, formerly Twitter, is seeking new investment at $44 billion - exactly what Elon Musk paid in 2022. Despite initial skepticism, the platform shows signs of revival. Morgan Stanley's $3 billion debt sale at face value signals growing investor confidence.
Meanwhile, Musk's empire flourishes. Tesla shares are up 40% since Trump's election. SpaceX now claims the title of world's largest tech startup at $350 billion. His AI venture, xAI, pursues funding at a potential $75 billion valuation.
This marks a dramatic shift from just months ago when Fidelity marked down its Twitter stake by 70%. While the investment round remains uncertain, it suggests a remarkable turnaround in market perception.
Why this matters:
- Musk's controversial leadership style might be working - at least financially. The original purchase price now looks less like an overpayment and more like a prediction.
- Social media platforms may be worth more in today's political landscape than traditional metrics suggest.
Read on, my dear:
- implicator.ai: From Twitter to X: The Billion-Dollar Comeback Story
- reuters: Musk's X in talks to raise money at a $44 billion valuation, Bloomberg News reports
- Bloomberg: Musk’s X Is in Talks to Raise Money at a $44 Billion Valuation
AI Photo of the Day
Prompt:
human eye blonde eyelashes man with greyish skin that’s tired
Battle of the Models: Meta Challenges DeepSeek with First AI Developer Conference
Meta is stepping into new territory. They're launching LlamaCon, their first developer conference focused purely on generative AI. It's happening April 29, and it's all about their Llama models.
The timing is interesting. Meta's been pushing hard on their "open" AI strategy, claiming hundreds of millions of model downloads. Big names like Goldman Sachs and AT&T are already using it. But there's a twist in this tale.
Enter DeepSeek, the Chinese AI company that's been giving Meta headaches. Word is, their latest model might outperform Meta's upcoming Llama release. Meta's response? Emergency war rooms and an $80 billion AI spending spree for 2024.
Mark Zuckerberg seems unfazed. He's promising new Llama models with "reasoning" abilities and something called "agentic" capabilities. That's fancy talk for AI that can think and act more independently. But there are storm clouds on the horizon. EU regulators are making Meta pump the brakes in some regions, and there's that pesky copyright lawsuit about book training data.
Why this matters:
- Meta's desperate sprint to catch up with DeepSeek reveals just how competitive the open-source AI race has become
- For developers, this power struggle could mean better tools - whether they come from Silicon Valley or Shanghai
- The real winner? Open source. When tech giants compete, innovation tends to accelerate
Read on, my dear:
- implicator.ai: Meta's LlamaCon: A Bold Bet on Open Source AI
AI Decoded 🔓
Your guide to mastering AI tools - no tech degree required.
🤖 Turn ChatGPT Into Your Lead Research Assistant
Want to stop manually researching leads? Here's how to automate it with Zapier and ChatGPT.
You'll need a Zapier account, ChatGPT API access, and your favorite CRM. That's it. ✨
The setup is straightforward. Create a Zap that triggers when a new lead enters your CRM. Connect it to ChatGPT to analyze the lead's company and potential. Then feed that intel back into your CRM.
Here's the meat of it: Tell ChatGPT to research each lead using this prompt:
Research: {{Lead Name}} at {{Company}}
Provide: Company description, industry, size
Rate: Sales potential (1-10)
Let the data flow back to your CRM. Map ChatGPT's output to your contact fields. Hit publish. 🎯 Your sales team now has its own AI research assistant. Every lead gets scored before anyone lifts a finger. And that 30 minutes you spent researching each lead? It's yours again. 🎉
AI & Tech News
OpenAI Study Shows AI's Coding Limitations
Even the most advanced AI models, including GPT-4 and Claude, can only successfully complete 26% of real-world coding tasks from Upwork. While they're quick at finding bugs, they struggle to understand why the bugs exist or how to fix them properly.
Intel stock soars on breakup rumor
Intel's stock surged 16.1% Tuesday following reports that Broadcom and Taiwan Semiconductor Manufacturing are considering bids to split up the company. The shares hit $27.39, marking their best performance since March 2020, as investors responded to rumors of Broadcom's interest in Intel's chip design division and TSMC's potential move for its manufacturing facilities.
From AI Pin to HP's Win: Humane's $116M Tech Sale
HP Inc. is acquiring Humane Inc.'s software platform, intellectual property, and most of its employees for $116 million, though notably excluding the troubled Ai Pin device business. The deal will see Humane's founders join HP to help integrate AI into computers and printers, while Humane's existing Ai Pin customers have until February 28 to download their data before server shutdown.
South Korea Unveils Plans for World's Largest AI Data Center
South Korea will build the world's largest AI data center, a $10 billion facility with 3-gigawatt capacity led by Fir Hills and backed by LG heir Brian Koo. The ambitious project in Jeollanam-do Province will create 10,000 jobs and transform South Korea into Asia's leading AI infrastructure hub when completed in 2028.
Gig Workers Face Sophie's Choice: Train Their AI Twin or Risk Obsolescence
Fiverr, the gig marketplace giant, is launching a controversial new service that lets freelancers create AI versions of themselves. For $25 monthly, workers can train AI models on their past work to automate future jobs, while retaining ownership of AI-generated content. CEO Micha Kaufman claims this will make freelancers "irreplaceable, not obsolete," but with gig work already declining due to AI competition, workers may feel pressured to participate or risk falling behind.
✨ Last But Not Least
Saving the best (or at least the most entertaining) bits for last - because good things come to those who scroll down.
When Smart Machines Get Stuck in the Digital Mud
Here's a plot twist for you: Our smartest AI models are getting worse at their jobs. Not because they're failing, but because they're thinking too hard.
A new study from UC Berkeley and ETH Zurich reveals an expensive irony. The more sophisticated we make our AI, the more likely it is to get lost in its own thoughts. Like that one colleague who spends two hours crafting the perfect email.
The numbers are sobering. Advanced AI models cost $1,400 per task, compared to simpler versions at $400. The performance difference? A measly 8%. We're essentially paying a premium for digital procrastination.
The researchers identified three classic overthinking patterns. There's "Analysis Paralysis" (endless planning), "Rogue Actions" (panic mode), and "Premature Disengagement" (giving up too soon). Sound familiar?
But there's hope. Simply choosing solutions with lower overthinking scores improved performance by 30% while cutting costs by 43%. Sometimes, the best strategy is just telling your AI to get on with it.
Why this matters:
- We've created artificial intelligence that's mastered humanity's finest trait: procrastination
- The most expensive solution isn't always the smartest – a lesson that applies to both silicon and carbon-based intelligence
- Sometimes, making things "smarter" just means giving them more ways to get stuck
Read on, my dear:
