- AI Fire
- Posts
- 🤖 Train O1-Preview Model for $450?
Plus: Edit 10,000 Images in One Click?
Read time: 5 minutes
In just 19 hours and for only $450, UC Berkeley developed an open-source model that rivals o1-preview, signaling a new era in AI reasoning.
What are on FIRE 🔥
PRESENTED BY ARTISAN
Hire Ava, the AI SDR & Get Meetings on Autopilot
Ava automates your entire outbound demand generation process, including:
Intent-Driven Lead Discovery
High Quality Emails with Waterfall Personalization
Follow-Up Management
Free up your sales team to focus on high-value interactions and closing deals, while Ava handles the time-consuming tasks.
AI INSIGHTS
🤖 Train Your Own O1-Preview Model for Under $450
AI reasoning models are locked behind closed doors—Sky-T1-32B-Preview just changed that.
The NovaSky team at UC Berkeley built a model that competes with o1-preview in reasoning and coding benchmarks. And they did it for just $450.
The $450 Build
NovaSky trained Sky-T1-32B-Preview on 8 H100 GPUs in just 19 hours, using DeepSpeed Zero-3 offload. Cost? $450 (Lambda Cloud pricing).
Key training steps:
Base Model: Fine-tuned Qwen2.5-32B-Instruct, an open-source model that lacked reasoning skills.
Training Data: 17K carefully curated reasoning samples (math, coding, science, puzzles).
Rejection Sampling: Tossed bad data. Verified math solutions. Ran coding unit tests.
Data Reformatting: Used GPT-4o-mini to rewrite traces, boosting accuracy from ~25% to over 90%.
What They Learned
Bigger models perform better. Smaller versions (7B, 14B) struggled with repetitive outputs.
Data balance is key. Too much math training hurt coding accuracy. A mix of hard math + coding data fixed it.
Reformatting helps. Without proper formatting, models misinterpret results. With reformatting, accuracy jumps.
What’s Next? Sky-T1-32B-Preview is just the start. NovaSky wants to:
Train more efficient models that are faster and cheaper.
Optimize reasoning logic for better performance in real-world tasks.
Get the open-source community involved to push AI reasoning forward.
Why it matters:
Big names like o1 and Gemini 2.0 dominate AI reasoning, but their models are proprietary black boxes. You can't access the training data, the model weights, or even the methods.
Sky-T1 flips that: 100% open-source—all code, data, and model weights are free to use. AI reasoning shouldn’t be locked behind closed doors. Sky-T1-32B-Preview proves you don’t need a billion-dollar budget to build something powerful.
Now, who's ready to train their own?
IN PARTNERSHIP WITH MASTERWORKS
Everyday investors net +$60M in proceeds from the sale of exclusive assets
From CEOs to shop owners, investors in Masterworks’ art offerings have received more than +$60,000,000 in total net proceeds to date (including principal) across their 23 exits.*
Surprised that so many people are interested in art investing? Bank of America recently found 83% of wealthy American investors 43 and under already collect, or want to. Normally, only the top 1% of investors would be able to diversify with art like Picassos and Banksys. But with Masterworks, you can easily diversify into this asset class without needing millions, or art expertise.
With a team that’s been working since 2019, Masterworks investors have realized representative annualized net returns like +17.6%, +17.8%, and +21.5% (among assets held for longer than one year).
The content is not intended to provide legal, tax, or investment advice.
No money is being solicited or will be accepted until the offering statement for a particular offering has been qualified by the SEC. Offers may be revoked at any time. Contacting Masterworks involves no commitment or obligation.
“Net Annualized Return” refers to the annualized internal rate of return net of all fees and expenses, calculated from the offering closing date to the date the sale is consummated. IRR may not be indicative of Masterworks paintings not yet sold and past performance is not indicative of future results. For additional information regarding the calculation of IRR for a particular investment in an artwork that has been sold, a reconciliation will be filed as an exhibit to Form 1-U and will be available on the SEC’s website. Masterworks has realized illustrative annualized net returns of 17.6% (1067 days held), 17.8% (672 days held), and 21.5% (638 days held) on 13 works held longer than one year (not inclusive of works held less than one year and unsold works).
“Net proceeds” represents the total liquidation proceeds distributed back to investors, net of all fees, expenses and proceeds reinvested in Masterworks offerings, of all works Masterworks has exited to date. This metric is not considered a presentation of performance but rather a mathematical figure that displays a platform metric on size, scale, and operation of the platform.
TODAY IN AI
AI HIGHLIGHTS
🔥 China is considering selling TikTok’s US operations to Musk if ByteDance loses its legal fight against the ban. Talks are early, and no deal is confirmed.
🚗 Google Cloud launched Automotive AI Agent, an advanced in-car assistant using Gemini AI. Mercedes-Benz will add it to MBUX Virtual Assistant in the new CLA, improving voice control and navigation.
🔍 The CMA (UK’s competition authority) is assessing if Google Search holds Strategic Market Status (SMS) under new digital market rules. Google argues its services boost the UK economy (£118B in 2023) and help businesses grow.
🎥 AI firms like OpenAI & Google are paying YouTubers & influencers for unused videos to train AI. Prices range from $1–$4 per minute. Creators earn extra, but copyright issues remain.
🤖 AI Assistants Will Have Eyes On Your Screens. Nvidia’s Project RTX brings a graphical AI assistant to Windows 11, watching your screen to help with tasks. Razer’s Project AVA gives real-time gaming tips to improve performance.
☑️ OpenAI just launched Tasks, letting paid ChatGPT users schedule reminders and recurring actions. This moves ChatGPT beyond chat into virtual assistant territory
DAILY AI FUNDRAISING
AI healthcare is booming! Qventus secured $105M to expand AI tools for hospitals, surgeries, and patient care. Its customer base grew 4x, and revenue tripled—pushing AI deeper into medical operations.
AI SOURCES FROM AI FIRE
🚀 Lead-Generating GPT Challenge – $549 Prize for Free
Something big is coming - and it’s completely free.
More than 2 years ago, before starting AI Fire, we was using AI just like everyone else. Then we learned how to customize it, and everything changed:
Automated the tasks I used to hate
Created AI-generated content that went viral (1.8M views)
Grew AI Fire’s newsletter to thousands of subscribers
Now, we are launching a 7-Day Lead-Generating GPT Challenge to help you do the same.
What’s in it for you?
A structured challenge to build real AI skills
A chance to win a $549 prize
No cost, no hidden fees—just practical learning
📅 Launching in 24 hours. Stay tuned.
📜 Get Your AI Fire Course Certificate
If you’ve completed an AI Fire course, you can now claim your personalized certificate for free.
How to get yours:
Finish the course
Fill out this form: Claim here
Receive your certificate by email
If you choose to share it on X, AI Fire will feature your achievement with a congratulatory post.
🏅 For AMCL Participants – Your Reward is Ready
If you completed the AI Mastermind Challenge League (AMCL) series and submitted feedback, check your email—you’ve been rewarded with one free month of AI Fire Academy 🔥🔥🔥.
This means you now have full access to all premium content at no cost.
If you haven’t received the email, check the address you used in the Feedback form. If there’s an issue, reach out to AI Fire support.
Enjoy your free access and keep learning.
NEW EMPOWERED AI TOOLS
📝 ResearchTrend.AI explores AI papers, join topic-based communities, and access visual insights.
✅ Theneo uploads once, get stunning and easy API docs.
🚀 Asendia AI personalized AI interviews that match your style.
🤖 TestSprite AI-powered testing for frontend and backend teams.
🌟 Astral automates marketing tasks with smart browser agents.
AI QUICK HITS
🎨 Adobe’s AI Can Edit 10,000 Images in One Click (Link)
🚀 Microsoft Launches CoreAI to Boost AI Development (Link)
🏥 Nvidia Partners with Healthcare Giants for AI Innovation (Link)
😱 French Woman Scammed $850K by AI Fake Brad Pitt (Link)
🏙️ A Look at Vatican City’s New Artificial Intelligence Law (Link)
AI CHART
AI is everywhere—more than half of Americans say they use it regularly. But while it’s making life easier, it’s also making people nervous.
The big question: Will AI create more jobs than it replaces?
A recent Ipsos survey asked people in 33 countries if they think AI will lead to many new jobs. The answers? A mix of optimism, doubt, and straight-up pessimism.
China is All In on AI Jobs
The most optimistic? China. No surprise there—government backing, booming AI startups, and a dominant academic research scene have put China at the forefront of AI expansion.
Other countries betting big on AI-driven job growth?
Indonesia and Thailand also rank near the top.
The U.S.? Not leading, but not at the bottom either.
Europe? Not So Sure.
On the other end of the spectrum, Hungary, Poland, and Germany aren’t buying the AI job hype.
Only 29% of Poles and Germans think AI will create new jobs.
Hungary? Just 24%.
Europe’s skepticism isn’t random—automation concerns, slow AI adoption, and regulatory hurdles make many doubt whether AI will actually add jobs rather than just replace them.
What This Means
Right now, the global outlook on AI and jobs is a tale of two mindsets:
Some countries see AI as an opportunity.
Others see it as a threat.
2025 will be the real test. Will AI prove the optimists right? Or will the skeptics be the ones saying, “Told you so”?
AI JOBS
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
*Disclaimer: AI Fire is a news publisher. The opinions in this content are from the authors or paid advertisers. The information shared is for general purposes only and is not financial advice. It should not be seen as an offer to buy or sell investments. AI Fire does not guarantee the accuracy of the information. You should do your own research and talk to a financial adviser before making any investments. The publisher and its affiliates are not responsible for any losses from using this information.
Reply