The Best Open-Source Model | Weekly Edition

PLUS HOT AI Tools & Tutorials

Dec 27, 2024

Hello and welcome to our weekly roundup!

OpenAI, Google, and Meta have already shared the best products, so this week, Chinese startup DeepSeek came on the scene and showed something outstanding. Many creators and reviewers believe the company has set a new benchmark as an open-source AI model!

Let's discuss this and much more in the last digest of the outgoing year!

In yesterday's post, we summarized our personal 2024 results and talked about the best AI tools of the year. It’s completely free, so check it out:
Year in AI | Summary & Tools of 2024
Stepan Ikaev
·
December 26, 2024
Read full story

This Creators’ AI Edition:

Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁

Your advertisement could be featured here!

Sponsor a spot in our newsletter to connect with subscribers interested in AI, technology, and startups. For inquiries, click the button below.

Featured Material 🎟️

The Best Open-Source AI Model by DeepSeek

The week's biggest news is the release of a new AI from Chinese startup DeepSeek. The company has introduced its latest big language model, DeepSeek-V3, now available as an open-source product. Developers can already download and modify this model under a permissive license. Initial tests show it's extremely good.

As not to bury you with details, here are the features you need to know:

Open-Source Availability: DeepSeek-V3 is fully open-source, allowing developers and researchers to access and adapt the model. This move aims to promote collaboration within the AI community.
Parameter and Architecture: The model features enormous 671B parameters and employs a Mixture-of-Experts (MoE) architecture that activates only 37B parameters for specific tasks. This design aims to improve efficiency and performance while lowering computational demands.
Performance Metrics: DeepSeek-V3 has been pre-trained on 14.8 trillion tokens and achieves a throughput of 60 tokens per second, which is three times faster than its predecessor, DeepSeek-V2.

In a subset of coding competitions hosted on Codeforces, DeepSeek outperforms Meta’s Llama 3.1 405B, OpenAI’s GPT-4o, and Alibaba’s Qwen 2.5 72B!

Task Specialization: The model demonstrates proficiency in technical tasks such as programming and mathematical problem-solving, scoring 92% on the HumanEval programming test and achieving 90.2 on the Math-500 challenge.

Keep your mailbox updated with key knowledge & news from the AI industry

In Artificial Analysis, which combines various tests, Deepseek-V3 scored 80 points. This puts it on par with heavyweights like Gemini 1.5 Pro and Claude Sonnet 3.5.

It also outperformed all other open-source models available today.

From a mathematical standpoint, DeepSeek-V3 is indeed impressive. However, if you include a little skepticism, you should note that larger models require more powerful hardware. For example, as TechCrunch notes, an unoptimized version of DeepSeek V3 would need a bank of high-end GPUs to answer questions at reasonable speeds.

The DeepSeek model is also customized to meet the requirements of the Chinese government. Therefore, some events are presented from a specific perspective. And sometimes, as in the case of Tiananmen Square, the AI will refuse to answer questions at all.

Right now, DeepSeek-V3 is available through Hugging Face and the official website. An API is offered to enterprises at promotional pricing until February 8, 2025.

News Of The Week 🌍

Results of 12 Days of OpenAI

The primary source of news during the whole month was 12 Days of OpenAI. This event gave us a lot of products and topics for discussion. That's why we have a separate post. You can find all the key announcements at the link below:

12 Days of OpenAI Recap

Stepan Ikaev

December 23, 2024

Read full story

ChatGPT Has Recovered After Second Outage

chatGPT stopped responding across all platforms and devices 12/26/24 - ChatGPT - OpenAI Developer Forum

ChatGPT experienced its second major outage on December 27th. According to DownDetector, which tracks outages in online services, users began reporting problems around 1:30 PM ET. Hours later, OpenAI updated its status page, saying that API and Sora were also experiencing problems.

At 11:00 PM ET, ChatGPT and Sora were back online.

This is the second time the chatbot has faced an outage. A similar event occurred on Dec. 11; access users also lost access to ChatGPT, Sora, and API. OpenAI later confirmed that the problem was with the upstream provider. Many also speculate that the company did not expect such product demand.

Microsoft & OpenAI Have Financial Definition of AGI

Apparently, Microsoft and OpenAI have their own definition of General Artificial Intelligence (AGI). And it doesn't look like what philosophers and tech experts mean by it. According to The Information, the companies signed an agreement last year that OpenAI would only reach AGI when its AI systems generate $100B in revenue. Given that the startup likely won't become profitable until 2029, even by that definition, AGI won't appear until the next decade.

I'd like to highlight another important detail in this story. Microsoft will lose access to OpenAI technology when the startup reaches that AGI.

Sharing is caring! Refer someone who recently started a learning journey in AI. Make them more productive and earn rewards!

Refer a friend

The Information: OpenAI Discusses Making Humanoid Robot

And another riverting inside from The Information. According to the media, OpenAI is considering building its humanoid robot. Details are scarce, but the report cites “two people with direct knowledge” about the project. OpenAI has previously invested in three startups, Figure, Physical Intelligence, and 1X, trying to realize similar ideas. Whether these companies will participate is unclear at this time.

Interestingly, OpenAI already had similar plans until 2021. However, a few years ago, the company decided to shut down the relevant division.

xAI Is Testing Grok's Standalone App for iOS

Grok now has a separate iOS app, though it is still in beta testing.

xAI, a startup founded by Elon Musk, is currently testing a standalone iOS app for its Grok chatbot. Previously available only through the X app, the Grok can now be accessed in Australia. The Grok app features similar capabilities to those offered in the X platform. You can utilize functions such as text rewriting, summarizing, question-and-answer assistance, and image generation from text prompts.

Additionally, Grok has the ability to access real-time data from both the web and X, enhancing its functionality.

Users have to create an account or link their existing X accounts to access the app.

Useful Tools ⚒️

Shoonya AI – Specialized foundation models fine-tuned for commerce use

LegalCheckPro – Turn legal documents into simple, clear summaries

EasyLang AI – Learn any language with AI in a fun, easy, personalized way

Picture Picker – Collect & categorize pictures 10x faster powered by AI

Lambda – Your AI-Powered Investment Copilot

Lambda is a portfolio management tool designed for individual investors. It uses AI, specifically GPT-4o, to provide real-time analysis and insights about your investments. This means you can chat with the AI to get personalized advice based on your portfolio. With this tool, you can track your holdings, manage dividends, and analyze risks. In addition, Lambda categorizes assets by stock type, sector, and country, helping you maintain a diversified portfolio.