☕️ OpenAI's o3-mini & Deep Research AI Agent Enter The Ring Against DeepSeek

DeepSeek may have stoked the fire with its R1 model last week, as OpenAI becomes one of the first to retaliate with the release of o3-mini and Deep Research. Other key highlights include:

• SoftBank and OpenAI launch a joint venture in Japan to market OpenAI's enterprise AI tech to Japan's biggest

• The Beatles win their 8th Grammy Award thanks to AI for bringing its final song to life

• The first compliance deadline arrives as the EU's AI Act goes into effect, banning AI systems with "unacceptable risk" levels

Join us at AI Tangle as we untangle this week's happenings in AI!Join us at AI Tangle as we untangle this week's happenings in AI!

OpenAI debuts o3-mini to become its most affordable frontier reasoning AI model

On Friday last week, OpenAI introduced o3-mini and o3-mini-high, the newest additions to its family of reasoning AI models, designed with STEM problem-solving in programming, math, and science in mind. Marketed as both "powerful" and "affordable" by OpenAI, o3-mini self-fact-checks to deliver more reliable results while offering faster response times and fewer major mistakes. Although o3-mini doesn't leapfrog ahead of DeepSeek's R1 or OpenAI's o1, the affordability of the model, along with limited free access for everyone, has made it an appealing powerhouse option.

​

​And they followed up on o3-mini with an AI agent called Deep Research​

However, OpenAI is adamant about staying in the spotlight - it released Deep Research, an AI agent designed to aid in complex and in-depth research, in a blog post on Sunday. To combat concerns regarding AI's accuracy in more complex subjects, OpenAI says it has mitigated many such problems by leveraging a fine-tuned, "special version" of o3. In doing so, OpenAI claims that Deep Research achieved an accuracy of 26.6% on the aptly named Humanity's Last Exam benchmark, about twice the accuracy of the runner-up, o3-mini-high at 13%. Unlike o3-mini, however, Deep Research doesn't come for free and is currently only available to Pro subscribers.

SoftBank and OpenAI launch a joint venture in Japan to market OpenAI's AI tech to Japan's biggest companies

During the "Transforming Business through AI" livestream event in Tokyo, Japanese investment giant SoftBank CEO Masayoshi Son and OpenAI CEO Sam Altman stepped up their partnership by announcing a new Japan AI joint venture. The two set up a 50:50 holding company called SB OpenAI Japan, which will market OpenAI's enterprise tech exclusively to major companies in Japan. Branding OpenAI's suite of AI tools as "Cristal Intelligence," SoftBank announced that it will spend $3 billion per year on this suite for itself and its subsidiaries, such as British chip designer Arm.

​

AI brings a decades-old Beatles song to life and wins the band its 8th Grammy Award

​Nearly 50 years after the band officially broke up, The Beatles' "Now and Then" won them the "Best Rock Performance" award at this year's Grammy's. "Now and Then" was supposed to be the band's final song, pieced together from demos spanning from the late-70s to the mid-90s. However, it was never released due to technical difficulties stopping Lennon's vocals and piano from being separated. Roughly 30 years later, Beatles members Ringo Starr and Paul McCartney, with the help of filmmaker Peter Jackson and his sound team, developed a machine-learning tool to split Lennon's voice from the piano, finishing the song and debuting it in 2023.​

​

EU's AI Act finally reaches its first compliance deadline, banning AI systems with "unacceptable risk" levels

Starting this Sunday, the European Union has reached a significant milestone as its regulators gain the power to ban AI systems posing "unacceptable risk" as the first compliance deadline on February 2 has hit. The EU's comprehensive AI Act categorizes applications into four risk levels - from minimal to unacceptable - with heavy fines for violations. Companies using prohibited applications may face fines of up to €35 million or 7% of annual revenue, though targeted law enforcement and specific therapeutic uses were brought out as exemptions.

​

Britain becomes the world's first nation to outlaw AI-generated child abuse tools

As the world's first, Britain is taking unprecedented steps to combat online child sexual abuse by making it illegal to use AI tools that generate explicit images, it announced on Saturday. The new law targets not only the creation but also the possession and distribution of these images, including the use of AI to "nudeify real-life images of children," the measures of which will be included in the Crime and Policing Bill when it comes to parliament. Authorities are also upping their investigative powers, while UK Interior Minister Yvette Cooper emphasized the connection between online activity and real-world abuse.

​

Tana emerges from stealth with a $14 million Series A fundraiser and an AI-native workspace

​Online workspace platform Tana recently emerged from stealth along with its new platform to announce a $14 million Series A round led by Tola Capital, bringing the startup's total funding to $25 million. The company's design, described as "a knowledge graph with connections that mimic the human brain," allows for rapid transformation of unstructured data into AI workflows using its Supertag feature. Early endorsements from industry leaders underscore Tana's bold vision for reshaping global team collaboration.

​Kusho - Kusho is an AI dev tool that leverages agents to test your user web journeys and inputs into a comprehensive ready-to-run test suite.

​QuillAI - Create and maintain high-ranking SEO-optimized content with QuillAI, an AI-powered content platform that saves time wasted on ineffective content.

​Cerebrella - Bring your sticky notes, whiteboard, research, visuals, and writing all under one AI-powered creative workspace umbrella with Cerebrella to foster brainstorming and capture ideas visually.

​Blaze - Blaze aims to make marketing simpler by using AI and automation to scan millions of online signals to help companies target the right potential customers (as soon as they show interest) and generate leads.

AI & The Future of Particle Physics at CERN (4-min read)

With AI becoming more prominent everywhere one can look, British physicist and 2026 CERN director Mark Thomson wants to bring more AI to the EU's premier particle physics lab to see if the tech can help scientists understand how the universe could end.

​Sam Altman's memo​ about AI agents "joining the workforce" in 2025 takes their first steps to become a reality with the release of Operator, the company's first major agent product. Other key takeaways of the week include:

• SoftBank, Oracle, and OpenAI announce a $500 billion US AI data center venture named Stargate Project

• Mistral CEO and co-founder Arthur Mensch confirms that the French AI startup is preparing for an IPO

• Another $1 billion from Google - The tech giant's investment in OpenAI rival Anthropic continues to grow

Join us at AI Tangle as we untangle this week's happenings in AI!

OpenAI takes its first major step into AI agent tech with the release of Operator

Not long ​after rumors of OpenAI's elusive AI agent nearing its launch​ began popping up, the AI poster child pulled the trigger and released its first attempt at changing the playing field - Operator. ​As detailed in the company's blog​, Operator is a general-purpose AI agent that can take control of a web browser and independently perform tasks on the web, all powered by OpenAI's Computer-Using Agent (CUA) model.

What more do we know of Operator?

The CUA model, specifically trained to interact with websites, combines the vision capabilities of GPT-4o and advanced reasoning from OpenAI's more advanced models, likely o1. Operator has been promised to automate tasks ranging from online shopping to reserving restaurants, though OpenAI warns that the CUA isn't perfect - it might struggle with more specialized tasks. Tasks like sending emails or banking transactions, however, require user supervision for security reasons.

Operator is currently out as a research preview, and only users in the US with the $200/mo Pro subscription plan will have access to it, though a rollout to Plus, Team, and Enterprise tiers will follow shortly. OpenAI CEO Sam Altman claimed Operator would be available in other countries soon enough, though European users specifically will have to wait a while.

SoftBank, Oracle, and OpenAI shake hands for a joint venture to pour $500 billion into US AI data centers

Earlier this week, ​the Stargate Project​ was unveiled at a press conference by SoftBank, Oracle, and OpenAI representatives at the White House, where President Donald Trump spoke at length about investment plans in US infrastructure. The goal is to pour $500 billion over four years into data centers in the US, starting with Texas, with an initial commitment of $100 billion, of which ​SoftBank and OpenAI have already pledged $19 billion each​. Additionally, Microsoft, Nvidia, and Arm are named as additional technical partners in what is one of the largest ventures in the US' AI sector.

French AI startup Mistral confirmed by company CEO Arthur Mensch to be setting up for an IPO

Following an ​interview with Bloomberg at the World Economic Forum​, Arthur Mensch, the co-founder and CEO of French AI startup Mistral, said that the company is preparing for an IPO while expanding into Asia-Pacific with a new office in Singapore. On top of that, Mensch also dismissed any acquisition rumors involving Microsoft, which, so far, has invested €15 million in the startup. Founded in April 2023 by former DeepMind and Meta researchers, Mistral has grown rapidly and raising $1.14 billion at a $6 billion valuation, with its most notable offerings including its ​Mistral Large​ and ​Le Chat​.

Google invests a further $1 billion in Silicon Valley AI unicorn Anthropic

Google is set to invest over $1 billion in AI startup Anthropic, separate from a recent $2 billion funding round led by Lightspeed Venture Partners, valuing the firm at $60 billion, ​according to an article by Financial Times​. This adds to ​Google's previous $2 billion worth of investment in the company​. Anthropic, a key rival of OpenAI, generates $875 million in annualized revenue through direct sales and partnerships like Amazon Web Services, through which it primarily sells access to enterprise clients to its models and technology.

Microsoft gets demoted from its position as OpenAI's exclusive cloud computing provider

On top of the Stargate news, Microsoft no longer finds itself as OpenAI's exclusive cloud provider but retains the "right of first refusal" for additional computing needs, ​as explained in a blog post by the tech giant​. Despite OpenAI's expanded partnerships and a recently made "new, large Azure commitment" for products and model training, Microsoft still holds onto exclusive API rights and IP usage for integration into their products, like Copilot.

ChatGPT goes down worldwide once again as OpenAI begins investigating the "increased errors"

ChatGPT has yet again experienced a significant outage on Thursday this week, with, as of writing, over 10,000 users in the UK alone reporting issues via Downdetector. Users encountered a "bad gateway error," and OpenAI has yet to provide an official explanation for the cause. The outage began at 11:00 GMT that day, disrupting services for many of the ​chatbot's 300 million weekly users​. While some users were understandably frustrated, ​others joked about the downtime on social media​ to highlight just how much they or others depend on it for everyday tasks.

Battery manufacturer SES secures $10 million in contracts to build AI-developed batteries

SES AI, a Li-Metal battery manufacturer, has recently secured contracts worth up to $10 million with two global OEM partners to develop AI-discovered materials for automotive batteries. These contracts build on SES's AI-powered Molecular Universe project, ​which recently delivered the first battery using an AI-discovered electrolyte​. Company CEO and founder Qichao Hu highlighted the contracts as a milestone for AI-driven battery innovation. SES' shares rose in accordance, up to 60% in premarket trading that day following the announcement.

​​Needle​ - Meet Needle, a seamlessly integrating platform that simplifies how teams access and manage knowledge across their favorite tools by leveraging AI to do organization-wide searches and save time.

​Transistor​ - Get the most out of your favorite podcasts with Transistor by automatically converting your audio into engaging, searchable, and accessible text.

​Minusx​ - Minusx seamlessly integrates into your analytics apps like Jupyter, Sheets, Metabase, and more, to automatically operate and handle your data for efficient and quick insights.

AI-Developed Drugs Are Coming Full Circle (​3-min read​)

Aside from coding, one area that AI has been increasingly more prominent is in healthcare, and at the World Economic Forum, Google DeepMind CEO Demis Hassabis talked in length about his enthusiasm, expecting AI-designed pharmaceuticals to be in clinical trials as early as the end of 2025.

How Far Away Really is Human-Level Intelligence? (​3-min read​)

Whether AI is overhyped or underhyped by one public figure or another, AI is advancing at rapid speeds. Although he's not calling it AGI, Anthropic CEO Dario Amodei believes the technology could overtake human capabilities not long after 2027.