Sam Altman: AGI coming in 2025

Hello, AI Enthusiasts!

Welcome to FavTutor’s AI Recap! We’ve gathered all the latest and important AI developments for the past 24 hours in one place, just for you.

In Today’s Newsletter: 😀 

• Sam Altman: AGI coming in 2025

• GPT-4o can now predict outputs of itself

• Google's ReCapture lets users change camera angles in existing videos

• OpenAI's new "Orion" model shows small gains over GPT-4

• Meta AI inks multi-year deal with Reuters for Real-time news

OpenAI
🧠 Sam Altman: AGI coming in 2025

While speaking with Garry Tan of Y Combinator, OpenAI CEO Sam Altman predicted the arrival of artificial general intelligence (AGI) in 2025. He said things are moving faster than expected and the path to AGI is "basically clear" See what he had to say in this quick, must-watch video! 👀

Insights for you:

• Artificial general Intelligence aims to develop software that can perform equally or better than humans on cognitive tasks. This is the point where the AI model is smart enough to reason, carry out tasks alone, create new ideas, and implement them.

• Altman believes engineering, not new science, will drive AGI's progress at OpenAI. Their researchers reportedly agree with Altman, despite slower gains in recent models.

• A few months ago, even Elon Musk said the same, predicting that an AGI will probably be smarter than a single human by 2025.

OpenAI
💡 GPT-4o can now predict outputs of itself

OpenAI has introduced a new feature called "Predicted Outputs" for its GPT-4o and GPT-4o-mini language models, making AI-supported text processing much faster. The feature is particularly suitable for repetitive tasks and minor document changes, but less so for completely new content.

Insights for you:

• Predicted Outputs is a new feature for the GPT-4o and GPT-4o-mini that accelerates AI-powered text processing.

• Response times are two to four times faster than existing models, with large file processing now taking around 20 seconds instead of 70.

• The system allows developers to pre-populate an expected portion of the output. Saving 50% of output tokens reduces latency by around 5%. Tests in Python, JavaScript, Go, and C++ have been successful.

Learn AI in 5 minutes a day.

The Rundown is the world’s most trusted AI newsletter, with over 700,000+ readers staying up-to-date with the latest AI news, understanding why it matters, and learning how to apply it in their work.

Their expert research team spends all day learning what’s new in AI, then distills the most important developments into one free email every morning.

Sign up.

Google
📸 Google's ReCapture lets users change camera angles in existing videos

Google researchers have developed ReCapture, a new AI technique that allows users to modify camera movements in videos after they've been recorded. The system aims to bring professional-grade video editing capabilities to casual users.

Insights for you:

• ReCapture is a method for retroactively adjusting video camerawork so that even non-professionals can perform professional post-production.

• ReCapture works in two steps: creating an anchor video with the desired new camera work, which may still contain errors, and then using masked video fine-tuning with temporal and spatial LoRA layers to improve the video synthesis.

• Rather than using an explicit 4D representation as an intermediate step, ReCapture taps into the motion knowledge stored in generative video models.

OpenAI
💪 OpenAI's new "Orion" model shows small gains over GPT-4

OpenAI's upcoming Orion model barely outperforms GPT-4, its predecessor, according to a new report from The Information. One reason for this is the lack of high-quality training data since most publicly available texts and data have already been used.

Insights for you:

• OpenAI's next major language model "Orion" has lower-than-expected performance improvements over its predecessor, GPT-4, especially in programming.

• Google's Gemini 2.0 is also said to fall short of expectations. Anthropic has yet to release its long-awaited Opus 3.5. The stagnation in the development of large-scale language models affects the entire AI industry.

• Despite the challenges, OpenAI CEO Sam Altman remains confident. The new North Star could be OpenAI's o1 model, which is expected to open a new scaling horizon by scaling computing power for inference instead of just training.

Meta
📰 Meta AI inks multi-year deal with Reuters for Real-time news

Under the deal, US users of Meta AI will get real-time access to Reuters news when asking questions about current events citing sources familiar with the matter, according to Axios. Answers to user queries about news will cite Reuters' stories and link out to its coverage.

Insights for you:

• Meta has signed a multi-year agreement with Reuters to integrate its content into the Meta AI chatbot.

• The deal puts Meta in competition with OpenAI, which has already signed several media partnerships for its ChatGPT chatbot, including with News Corp, Vox Media, The Atlantic, TIME, and the Financial Times.

• Tech companies' unregulated media deals are problematic because they can selectively decide what news is selected and distributed. They threaten media diversity.