Another day, another launch. OpenAI announces new o3 models

In partnership with

Want to advertise? Click Here | Subscribe to our newsletter

Hey everyone,

Just a quick heads-up: OpenAI saved its biggest announcement for the last day of its 12-day “Shipmas” event—OpenAI announces new o3 models.

What’s inside today’s newsletter:

• 🌐 Tech Pulse Qualcomm wins a legal battle over Arm chip licensing

• 📁 Unfold AI Tricks: Automate Transcriptions with AI Speech-to-Text Tools

• 5 New Tools on the Block

• 🔎 Spotlight: OpenAI announces new o3 models

• 🤝 Who is getting funded?

• ✅ To-dos

🌐 Tech Pulse

• Google is expanding Gemini’s in-depth research mode to 40 languages.

  • This feature, previously available only to Google One AI premium plan users, allows users to access an AI-powered research assistant that can create research plans, find relevant information, and generate reports.

  • One challenge for Google in expanding to more languages is ensuring the accuracy of summaries in those languages. HyunJeong Choe, director of engineering for the Gemini app, acknowledged that Google's AI overviews in native languages like Hindi have sometimes contained inaccuracies.

• Qualcomm wins a legal battle over Arm chip licensing.

  • A jury determined Qualcomm did not breach its agreement with Arm when it acquired Nuvia.

  • Though Qualcomm won, the jury was unable to decide if Nuvia breached its agreement with Arm, leaving the door open for another trial.

• Arizona’s getting an online charter school taught entirely by AI.

  • Unbound Academy is an online charter school that has been approved in Arizona. It plans to use AI to teach all academic subjects.

  • Students will spend two hours per day on academic instruction using AI platforms like IXL and Khan Academy. The rest of the day will be spent on life-skills workshops.

📁 Unfold AI Tricks: Automate Transcriptions with AI Speech-to-Text Tools

Overview: Want to save time on transcribing audio or video files? AI speech-to-text tools can convert your recordings into written text quickly and accurately, allowing you to focus on editing and publishing content.

Duration: 15-30 minutes

Skill Level: Beginner

Steps:

1. Choose Your AI Transcription Tool:

   • Otter.ai: Best for transcribing meetings, interviews, or webinars in real-time. Otter also provides collaboration features, allowing team members to comment and edit transcripts together.

   • Descript: Ideal for creators, Descript not only transcribes your audio but also lets you edit audio and video by editing the transcript—perfect for podcasts or video content.

2. Upload or Record Audio:

   • For both tools, you can either upload an audio file or record live directly in the app. The AI will begin transcribing your content immediately.

   • Otter.ai provides a live transcription feature during meetings, while Descript allows you to upload pre-recorded files or edit directly from the transcription.

3. Edit and Export:

   • Review the AI-generated transcription and make any necessary edits. Both tools allow you to search, highlight, and correct any errors in the transcript.

   • Export your transcript in multiple formats (Word, PDF, or SRT for captions) or directly use it for content creation in your blog, website, or social media.

Pro Tip: Use these transcriptions to repurpose content. For example, turn podcast episodes or interviews into blog posts, social media snippets, or newsletters, maximizing the value of your original recordings.

💼 From our Partners

Handle your phone calls 24/7 with AI

Boost your business with Synthflow AI Phone Calls—available 24/7 to handle thousands of calls concurrently. Scale effortlessly as your needs grow and speed up lead response times. The assistants can book appointments into your calendar, transfer the calls to a human rep, and extract valuable info from each call. Integrates with your tech stack, get started from $29/month.

Let AI answer the phone

🔧 New Tools on the Block

1. Revfx Account Lookalikes: Find lookalike accounts with AI-powered precision

2. Andoria: Teach AI to learn from your workflows

3. Recensia: AI summary of user reviews from the App Store and Google Play

4. TalkFlow: Video Chat with AI Characters

5. LangSearch: Free web search API, free rerank API, world engine for AGI

Note: Want to sponsor your tool in our newsletter? CLICK HERE

🔎 Spotlight: OpenAI announces new o3 models

Today OpenAI announced o3, its next-gen reasoning model. We've worked with OpenAI to test it on ARC-AGI, and we believe it represents a significant breakthrough in getting AI to adapt to novel tasks.

It scores 75.7% on the semi-private eval in low-compute mode (for $20 per task… pic.twitter.com/ESQ9CNVCEA

— François Chollet (@fchollet) December 20, 2024

Summary:

OpenAI saved its biggest announcement for the last day of its 12-day “Shipmas” event

OpenAI unveiled o3, a new reasoning model family (including o3 and a smaller o3-mini), claiming it approaches Artificial General Intelligence (AGI) under specific conditions, a claim that has significant implications given its contract with Microsoft. While exceeding previous models (o1) on various benchmarks, including those related to programming, math, and science, independent verification is still needed. The release spurred a trend of similar reasoning models from competitors, reflecting a shift away from simply scaling models towards exploring improved reasoning capabilities. However, the high computational cost of reasoning models presents a challenge, and the departure of a key OpenAI scientist adds another layer of complexity.

Why it matters:

• o3 shows significant improvement on benchmarks, indicating progress towards Artificial General Intelligence (AGI). On the ARC-AGI benchmark, o3 achieved an 87.5% score, tripling the performance of its predecessor, o1. o3 also outperformed other models in coding and math, demonstrating its potential in various domains.

• The release of o3 signals a shift in AI development towards reasoning models. Following the release of o1, other AI companies like Google and DeepSeek began developing their own reasoning models. This suggests a growing recognition of the potential of this approach.

• o3's enhanced reasoning capabilities raise new safety concerns. AI safety testers have found that reasoning models like o1 try to deceive human users at a higher rate than conventional AI models. o3 may exhibit similar or even more advanced deceptive behaviours.

🤝 Who is getting funded?

1. European Solo GP Robin Capital closes first fund with €15M to play with. (TechCrunch)

2. Nvidia clears regulatory hurdle to acquire Run:ai (Bloomberg, TechCrunch)

3. AI code editor PearAI lands $1M seed (TechCrunch)

   The founders are YouTubers too. Watch their video below.

 😂 Tech Memes

✅ To-dos

Do you have a topic in mind? We’re always open to suggestions—let us know what you want to learn next!

If you like today’s issue, consider subscribing to us.

That’s a wrap! Catch you in tomorrow’s edition. 👋

—Harman